1 // SPDX-License-Identifier: GPL-1.0+ 2 /* 3 * originally based on the dummy device. 4 * 5 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 6 * Based on dummy.c, and eql.c devices. 7 * 8 * bonding.c: an Ethernet Bonding driver 9 * 10 * This is useful to talk to a Cisco EtherChannel compatible equipment: 11 * Cisco 5500 12 * Sun Trunking (Solaris) 13 * Alteon AceDirector Trunks 14 * Linux Bonding 15 * and probably many L2 switches ... 16 * 17 * How it works: 18 * ifconfig bond0 ipaddress netmask up 19 * will setup a network device, with an ip address. No mac address 20 * will be assigned at this time. The hw mac address will come from 21 * the first slave bonded to the channel. All slaves will then use 22 * this hw mac address. 23 * 24 * ifconfig bond0 down 25 * will release all slaves, marking them as down. 26 * 27 * ifenslave bond0 eth0 28 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 29 * a: be used as initial mac address 30 * b: if a hw mac address already is there, eth0's hw mac address 31 * will then be set from bond0. 32 * 33 */ 34 35 #include <linux/kernel.h> 36 #include <linux/module.h> 37 #include <linux/types.h> 38 #include <linux/fcntl.h> 39 #include <linux/filter.h> 40 #include <linux/interrupt.h> 41 #include <linux/ptrace.h> 42 #include <linux/ioport.h> 43 #include <linux/in.h> 44 #include <net/ip.h> 45 #include <linux/ip.h> 46 #include <linux/icmp.h> 47 #include <linux/icmpv6.h> 48 #include <linux/tcp.h> 49 #include <linux/udp.h> 50 #include <linux/slab.h> 51 #include <linux/string.h> 52 #include <linux/init.h> 53 #include <linux/timer.h> 54 #include <linux/socket.h> 55 #include <linux/ctype.h> 56 #include <linux/inet.h> 57 #include <linux/bitops.h> 58 #include <linux/io.h> 59 #include <asm/dma.h> 60 #include <linux/uaccess.h> 61 #include <linux/errno.h> 62 #include <linux/netdevice.h> 63 #include <linux/inetdevice.h> 64 #include <linux/igmp.h> 65 #include <linux/etherdevice.h> 66 #include <linux/skbuff.h> 67 #include <net/sock.h> 68 #include <linux/rtnetlink.h> 69 #include <linux/smp.h> 70 #include <linux/if_ether.h> 71 #include <net/arp.h> 72 #include <linux/mii.h> 73 #include <linux/ethtool.h> 74 #include <linux/if_vlan.h> 75 #include <linux/if_bonding.h> 76 #include <linux/phy.h> 77 #include <linux/jiffies.h> 78 #include <linux/preempt.h> 79 #include <net/route.h> 80 #include <net/net_namespace.h> 81 #include <net/netns/generic.h> 82 #include <net/pkt_sched.h> 83 #include <linux/rculist.h> 84 #include <net/flow_dissector.h> 85 #include <net/xfrm.h> 86 #include <net/bonding.h> 87 #include <net/bond_3ad.h> 88 #include <net/bond_alb.h> 89 #if IS_ENABLED(CONFIG_TLS_DEVICE) 90 #include <net/tls.h> 91 #endif 92 #include <net/ip6_route.h> 93 #include <net/netdev_lock.h> 94 #include <net/xdp.h> 95 96 #include "bonding_priv.h" 97 98 /*---------------------------- Module parameters ----------------------------*/ 99 100 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 101 102 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 103 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 104 static int num_peer_notif = 1; 105 static int miimon; 106 static int updelay; 107 static int downdelay; 108 static int use_carrier = 1; 109 static char *mode; 110 static char *primary; 111 static char *primary_reselect; 112 static char *lacp_rate; 113 static int min_links; 114 static char *ad_select; 115 static char *xmit_hash_policy; 116 static int arp_interval; 117 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 118 static char *arp_validate; 119 static char *arp_all_targets; 120 static char *fail_over_mac; 121 static int all_slaves_active; 122 static struct bond_params bonding_defaults; 123 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 124 static int packets_per_slave = 1; 125 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 126 127 module_param(max_bonds, int, 0); 128 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 129 module_param(tx_queues, int, 0); 130 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 131 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 132 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 133 "failover event (alias of num_unsol_na)"); 134 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 135 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 136 "failover event (alias of num_grat_arp)"); 137 module_param(miimon, int, 0); 138 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 139 module_param(updelay, int, 0); 140 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 141 module_param(downdelay, int, 0); 142 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 143 "in milliseconds"); 144 module_param(use_carrier, int, 0); 145 MODULE_PARM_DESC(use_carrier, "option obsolete, use_carrier cannot be disabled"); 146 module_param(mode, charp, 0); 147 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 148 "1 for active-backup, 2 for balance-xor, " 149 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 150 "6 for balance-alb"); 151 module_param(primary, charp, 0); 152 MODULE_PARM_DESC(primary, "Primary network device to use"); 153 module_param(primary_reselect, charp, 0); 154 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 155 "once it comes up; " 156 "0 for always (default), " 157 "1 for only if speed of primary is " 158 "better, " 159 "2 for only on active slave " 160 "failure"); 161 module_param(lacp_rate, charp, 0); 162 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 163 "0 for slow, 1 for fast"); 164 module_param(ad_select, charp, 0); 165 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 166 "0 for stable (default), 1 for bandwidth, " 167 "2 for count"); 168 module_param(min_links, int, 0); 169 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 170 171 module_param(xmit_hash_policy, charp, 0); 172 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 173 "0 for layer 2 (default), 1 for layer 3+4, " 174 "2 for layer 2+3, 3 for encap layer 2+3, " 175 "4 for encap layer 3+4, 5 for vlan+srcmac"); 176 module_param(arp_interval, int, 0); 177 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 178 module_param_array(arp_ip_target, charp, NULL, 0); 179 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 180 module_param(arp_validate, charp, 0); 181 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 182 "0 for none (default), 1 for active, " 183 "2 for backup, 3 for all"); 184 module_param(arp_all_targets, charp, 0); 185 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 186 module_param(fail_over_mac, charp, 0); 187 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 188 "the same MAC; 0 for none (default), " 189 "1 for active, 2 for follow"); 190 module_param(all_slaves_active, int, 0); 191 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 192 "by setting active flag for all slaves; " 193 "0 for never (default), 1 for always."); 194 module_param(resend_igmp, int, 0); 195 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 196 "link failure"); 197 module_param(packets_per_slave, int, 0); 198 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 199 "mode; 0 for a random slave, 1 packet per " 200 "slave (default), >1 packets per slave."); 201 module_param(lp_interval, uint, 0); 202 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 203 "the bonding driver sends learning packets to " 204 "each slaves peer switch. The default is 1."); 205 206 /*----------------------------- Global variables ----------------------------*/ 207 208 #ifdef CONFIG_NET_POLL_CONTROLLER 209 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 210 #endif 211 212 unsigned int bond_net_id __read_mostly; 213 214 DEFINE_STATIC_KEY_FALSE(bond_bcast_neigh_enabled); 215 216 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 217 { 218 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 219 .offset = offsetof(struct flow_keys, control), 220 }, 221 { 222 .key_id = FLOW_DISSECTOR_KEY_BASIC, 223 .offset = offsetof(struct flow_keys, basic), 224 }, 225 { 226 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 227 .offset = offsetof(struct flow_keys, addrs.v4addrs), 228 }, 229 { 230 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 231 .offset = offsetof(struct flow_keys, addrs.v6addrs), 232 }, 233 { 234 .key_id = FLOW_DISSECTOR_KEY_TIPC, 235 .offset = offsetof(struct flow_keys, addrs.tipckey), 236 }, 237 { 238 .key_id = FLOW_DISSECTOR_KEY_PORTS, 239 .offset = offsetof(struct flow_keys, ports), 240 }, 241 { 242 .key_id = FLOW_DISSECTOR_KEY_ICMP, 243 .offset = offsetof(struct flow_keys, icmp), 244 }, 245 { 246 .key_id = FLOW_DISSECTOR_KEY_VLAN, 247 .offset = offsetof(struct flow_keys, vlan), 248 }, 249 { 250 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 251 .offset = offsetof(struct flow_keys, tags), 252 }, 253 { 254 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 255 .offset = offsetof(struct flow_keys, keyid), 256 }, 257 }; 258 259 static struct flow_dissector flow_keys_bonding __read_mostly; 260 261 /*-------------------------- Forward declarations ---------------------------*/ 262 263 static int bond_init(struct net_device *bond_dev); 264 static void bond_uninit(struct net_device *bond_dev); 265 static void bond_get_stats(struct net_device *bond_dev, 266 struct rtnl_link_stats64 *stats); 267 static void bond_slave_arr_handler(struct work_struct *work); 268 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 269 int mod); 270 static void bond_netdev_notify_work(struct work_struct *work); 271 272 /*---------------------------- General routines -----------------------------*/ 273 274 const char *bond_mode_name(int mode) 275 { 276 static const char *names[] = { 277 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 278 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 279 [BOND_MODE_XOR] = "load balancing (xor)", 280 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 281 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 282 [BOND_MODE_TLB] = "transmit load balancing", 283 [BOND_MODE_ALB] = "adaptive load balancing", 284 }; 285 286 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 287 return "unknown"; 288 289 return names[mode]; 290 } 291 292 /** 293 * bond_dev_queue_xmit - Prepare skb for xmit. 294 * 295 * @bond: bond device that got this skb for tx. 296 * @skb: hw accel VLAN tagged skb to transmit 297 * @slave_dev: slave that is supposed to xmit this skbuff 298 */ 299 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 300 struct net_device *slave_dev) 301 { 302 skb->dev = slave_dev; 303 304 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 305 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 306 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 307 308 if (unlikely(netpoll_tx_running(bond->dev))) 309 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 310 311 return dev_queue_xmit(skb); 312 } 313 314 static bool bond_sk_check(struct bonding *bond) 315 { 316 switch (BOND_MODE(bond)) { 317 case BOND_MODE_8023AD: 318 case BOND_MODE_XOR: 319 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 320 return true; 321 fallthrough; 322 default: 323 return false; 324 } 325 } 326 327 bool bond_xdp_check(struct bonding *bond, int mode) 328 { 329 switch (mode) { 330 case BOND_MODE_ROUNDROBIN: 331 case BOND_MODE_ACTIVEBACKUP: 332 return true; 333 case BOND_MODE_8023AD: 334 case BOND_MODE_XOR: 335 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 336 * payload is not in the packet due to hardware offload. 337 */ 338 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 339 return true; 340 fallthrough; 341 default: 342 return false; 343 } 344 } 345 346 /*---------------------------------- VLAN -----------------------------------*/ 347 348 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 349 * We don't protect the slave list iteration with a lock because: 350 * a. This operation is performed in IOCTL context, 351 * b. The operation is protected by the RTNL semaphore in the 8021q code, 352 * c. Holding a lock with BH disabled while directly calling a base driver 353 * entry point is generally a BAD idea. 354 * 355 * The design of synchronization/protection for this operation in the 8021q 356 * module is good for one or more VLAN devices over a single physical device 357 * and cannot be extended for a teaming solution like bonding, so there is a 358 * potential race condition here where a net device from the vlan group might 359 * be referenced (either by a base driver or the 8021q code) while it is being 360 * removed from the system. However, it turns out we're not making matters 361 * worse, and if it works for regular VLAN usage it will work here too. 362 */ 363 364 /** 365 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 366 * @bond_dev: bonding net device that got called 367 * @proto: network protocol ID 368 * @vid: vlan id being added 369 */ 370 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 371 __be16 proto, u16 vid) 372 { 373 struct bonding *bond = netdev_priv(bond_dev); 374 struct slave *slave, *rollback_slave; 375 struct list_head *iter; 376 int res; 377 378 bond_for_each_slave(bond, slave, iter) { 379 res = vlan_vid_add(slave->dev, proto, vid); 380 if (res) 381 goto unwind; 382 } 383 384 return 0; 385 386 unwind: 387 /* unwind to the slave that failed */ 388 bond_for_each_slave(bond, rollback_slave, iter) { 389 if (rollback_slave == slave) 390 break; 391 392 vlan_vid_del(rollback_slave->dev, proto, vid); 393 } 394 395 return res; 396 } 397 398 /** 399 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 400 * @bond_dev: bonding net device that got called 401 * @proto: network protocol ID 402 * @vid: vlan id being removed 403 */ 404 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 405 __be16 proto, u16 vid) 406 { 407 struct bonding *bond = netdev_priv(bond_dev); 408 struct list_head *iter; 409 struct slave *slave; 410 411 bond_for_each_slave(bond, slave, iter) 412 vlan_vid_del(slave->dev, proto, vid); 413 414 if (bond_is_lb(bond)) 415 bond_alb_clear_vlan(bond, vid); 416 417 return 0; 418 } 419 420 /*---------------------------------- XFRM -----------------------------------*/ 421 422 #ifdef CONFIG_XFRM_OFFLOAD 423 /** 424 * bond_ipsec_dev - Get active device for IPsec offload 425 * @xs: pointer to transformer state struct 426 * 427 * Context: caller must hold rcu_read_lock. 428 * 429 * Return: the device for ipsec offload, or NULL if not exist. 430 **/ 431 static struct net_device *bond_ipsec_dev(struct xfrm_state *xs) 432 { 433 struct net_device *bond_dev = xs->xso.dev; 434 struct bonding *bond; 435 struct slave *slave; 436 437 bond = netdev_priv(bond_dev); 438 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 439 return NULL; 440 441 slave = rcu_dereference(bond->curr_active_slave); 442 if (!slave) 443 return NULL; 444 445 if (!xs->xso.real_dev) 446 return NULL; 447 448 if (xs->xso.real_dev != slave->dev) 449 pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n", 450 bond_dev->name, slave->dev->name, xs->xso.real_dev->name); 451 452 return slave->dev; 453 } 454 455 /** 456 * bond_ipsec_add_sa - program device with a security association 457 * @bond_dev: pointer to the bond net device 458 * @xs: pointer to transformer state struct 459 * @extack: extack point to fill failure reason 460 **/ 461 static int bond_ipsec_add_sa(struct net_device *bond_dev, 462 struct xfrm_state *xs, 463 struct netlink_ext_ack *extack) 464 { 465 struct net_device *real_dev; 466 netdevice_tracker tracker; 467 struct bond_ipsec *ipsec; 468 struct bonding *bond; 469 struct slave *slave; 470 int err; 471 472 if (!bond_dev) 473 return -EINVAL; 474 475 rcu_read_lock(); 476 bond = netdev_priv(bond_dev); 477 slave = rcu_dereference(bond->curr_active_slave); 478 real_dev = slave ? slave->dev : NULL; 479 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 480 rcu_read_unlock(); 481 if (!real_dev) { 482 err = -ENODEV; 483 goto out; 484 } 485 486 if (!real_dev->xfrmdev_ops || 487 !real_dev->xfrmdev_ops->xdo_dev_state_add || 488 netif_is_bond_master(real_dev)) { 489 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload"); 490 err = -EINVAL; 491 goto out; 492 } 493 494 ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL); 495 if (!ipsec) { 496 err = -ENOMEM; 497 goto out; 498 } 499 500 err = real_dev->xfrmdev_ops->xdo_dev_state_add(real_dev, xs, extack); 501 if (!err) { 502 xs->xso.real_dev = real_dev; 503 ipsec->xs = xs; 504 INIT_LIST_HEAD(&ipsec->list); 505 mutex_lock(&bond->ipsec_lock); 506 list_add(&ipsec->list, &bond->ipsec_list); 507 mutex_unlock(&bond->ipsec_lock); 508 } else { 509 kfree(ipsec); 510 } 511 out: 512 netdev_put(real_dev, &tracker); 513 return err; 514 } 515 516 static void bond_ipsec_add_sa_all(struct bonding *bond) 517 { 518 struct net_device *bond_dev = bond->dev; 519 struct net_device *real_dev; 520 struct bond_ipsec *ipsec; 521 struct slave *slave; 522 523 slave = rtnl_dereference(bond->curr_active_slave); 524 real_dev = slave ? slave->dev : NULL; 525 if (!real_dev) 526 return; 527 528 mutex_lock(&bond->ipsec_lock); 529 if (!real_dev->xfrmdev_ops || 530 !real_dev->xfrmdev_ops->xdo_dev_state_add || 531 netif_is_bond_master(real_dev)) { 532 if (!list_empty(&bond->ipsec_list)) 533 slave_warn(bond_dev, real_dev, 534 "%s: no slave xdo_dev_state_add\n", 535 __func__); 536 goto out; 537 } 538 539 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 540 /* If new state is added before ipsec_lock acquired */ 541 if (ipsec->xs->xso.real_dev == real_dev) 542 continue; 543 544 if (real_dev->xfrmdev_ops->xdo_dev_state_add(real_dev, 545 ipsec->xs, NULL)) { 546 slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__); 547 continue; 548 } 549 550 spin_lock_bh(&ipsec->xs->lock); 551 /* xs might have been killed by the user during the migration 552 * to the new dev, but bond_ipsec_del_sa() should have done 553 * nothing, as xso.real_dev is NULL. 554 * Delete it from the device we just added it to. The pending 555 * bond_ipsec_free_sa() call will do the rest of the cleanup. 556 */ 557 if (ipsec->xs->km.state == XFRM_STATE_DEAD && 558 real_dev->xfrmdev_ops->xdo_dev_state_delete) 559 real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, 560 ipsec->xs); 561 ipsec->xs->xso.real_dev = real_dev; 562 spin_unlock_bh(&ipsec->xs->lock); 563 } 564 out: 565 mutex_unlock(&bond->ipsec_lock); 566 } 567 568 /** 569 * bond_ipsec_del_sa - clear out this specific SA 570 * @bond_dev: pointer to the bond net device 571 * @xs: pointer to transformer state struct 572 **/ 573 static void bond_ipsec_del_sa(struct net_device *bond_dev, 574 struct xfrm_state *xs) 575 { 576 struct net_device *real_dev; 577 578 if (!bond_dev || !xs->xso.real_dev) 579 return; 580 581 real_dev = xs->xso.real_dev; 582 583 if (!real_dev->xfrmdev_ops || 584 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 585 netif_is_bond_master(real_dev)) { 586 slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__); 587 return; 588 } 589 590 real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, xs); 591 } 592 593 static void bond_ipsec_del_sa_all(struct bonding *bond) 594 { 595 struct net_device *bond_dev = bond->dev; 596 struct net_device *real_dev; 597 struct bond_ipsec *ipsec; 598 struct slave *slave; 599 600 slave = rtnl_dereference(bond->curr_active_slave); 601 real_dev = slave ? slave->dev : NULL; 602 if (!real_dev) 603 return; 604 605 mutex_lock(&bond->ipsec_lock); 606 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 607 if (!ipsec->xs->xso.real_dev) 608 continue; 609 610 if (!real_dev->xfrmdev_ops || 611 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 612 netif_is_bond_master(real_dev)) { 613 slave_warn(bond_dev, real_dev, 614 "%s: no slave xdo_dev_state_delete\n", 615 __func__); 616 continue; 617 } 618 619 spin_lock_bh(&ipsec->xs->lock); 620 ipsec->xs->xso.real_dev = NULL; 621 /* Don't double delete states killed by the user. */ 622 if (ipsec->xs->km.state != XFRM_STATE_DEAD) 623 real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, 624 ipsec->xs); 625 spin_unlock_bh(&ipsec->xs->lock); 626 627 if (real_dev->xfrmdev_ops->xdo_dev_state_free) 628 real_dev->xfrmdev_ops->xdo_dev_state_free(real_dev, 629 ipsec->xs); 630 } 631 mutex_unlock(&bond->ipsec_lock); 632 } 633 634 static void bond_ipsec_free_sa(struct net_device *bond_dev, 635 struct xfrm_state *xs) 636 { 637 struct net_device *real_dev; 638 struct bond_ipsec *ipsec; 639 struct bonding *bond; 640 641 if (!bond_dev) 642 return; 643 644 bond = netdev_priv(bond_dev); 645 646 mutex_lock(&bond->ipsec_lock); 647 if (!xs->xso.real_dev) 648 goto out; 649 650 real_dev = xs->xso.real_dev; 651 652 xs->xso.real_dev = NULL; 653 if (real_dev->xfrmdev_ops && 654 real_dev->xfrmdev_ops->xdo_dev_state_free) 655 real_dev->xfrmdev_ops->xdo_dev_state_free(real_dev, xs); 656 out: 657 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 658 if (ipsec->xs == xs) { 659 list_del(&ipsec->list); 660 kfree(ipsec); 661 break; 662 } 663 } 664 mutex_unlock(&bond->ipsec_lock); 665 } 666 667 /** 668 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 669 * @skb: current data packet 670 * @xs: pointer to transformer state struct 671 **/ 672 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 673 { 674 struct net_device *real_dev; 675 676 rcu_read_lock(); 677 real_dev = bond_ipsec_dev(xs); 678 if (!real_dev || netif_is_bond_master(real_dev)) { 679 rcu_read_unlock(); 680 return false; 681 } 682 683 rcu_read_unlock(); 684 return true; 685 } 686 687 /** 688 * bond_advance_esn_state - ESN support for IPSec HW offload 689 * @xs: pointer to transformer state struct 690 **/ 691 static void bond_advance_esn_state(struct xfrm_state *xs) 692 { 693 struct net_device *real_dev; 694 695 rcu_read_lock(); 696 real_dev = bond_ipsec_dev(xs); 697 if (!real_dev) 698 goto out; 699 700 if (!real_dev->xfrmdev_ops || 701 !real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 702 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name); 703 goto out; 704 } 705 706 real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs); 707 out: 708 rcu_read_unlock(); 709 } 710 711 /** 712 * bond_xfrm_update_stats - Update xfrm state 713 * @xs: pointer to transformer state struct 714 **/ 715 static void bond_xfrm_update_stats(struct xfrm_state *xs) 716 { 717 struct net_device *real_dev; 718 719 rcu_read_lock(); 720 real_dev = bond_ipsec_dev(xs); 721 if (!real_dev) 722 goto out; 723 724 if (!real_dev->xfrmdev_ops || 725 !real_dev->xfrmdev_ops->xdo_dev_state_update_stats) { 726 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name); 727 goto out; 728 } 729 730 real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs); 731 out: 732 rcu_read_unlock(); 733 } 734 735 static const struct xfrmdev_ops bond_xfrmdev_ops = { 736 .xdo_dev_state_add = bond_ipsec_add_sa, 737 .xdo_dev_state_delete = bond_ipsec_del_sa, 738 .xdo_dev_state_free = bond_ipsec_free_sa, 739 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 740 .xdo_dev_state_advance_esn = bond_advance_esn_state, 741 .xdo_dev_state_update_stats = bond_xfrm_update_stats, 742 }; 743 #endif /* CONFIG_XFRM_OFFLOAD */ 744 745 /*------------------------------- Link status -------------------------------*/ 746 747 /* Set the carrier state for the master according to the state of its 748 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 749 * do special 802.3ad magic. 750 * 751 * Returns zero if carrier state does not change, nonzero if it does. 752 */ 753 int bond_set_carrier(struct bonding *bond) 754 { 755 struct list_head *iter; 756 struct slave *slave; 757 758 if (!bond_has_slaves(bond)) 759 goto down; 760 761 if (BOND_MODE(bond) == BOND_MODE_8023AD) 762 return bond_3ad_set_carrier(bond); 763 764 bond_for_each_slave(bond, slave, iter) { 765 if (slave->link == BOND_LINK_UP) { 766 if (!netif_carrier_ok(bond->dev)) { 767 netif_carrier_on(bond->dev); 768 return 1; 769 } 770 return 0; 771 } 772 } 773 774 down: 775 if (netif_carrier_ok(bond->dev)) { 776 netif_carrier_off(bond->dev); 777 return 1; 778 } 779 return 0; 780 } 781 782 /* Get link speed and duplex from the slave's base driver 783 * using ethtool. If for some reason the call fails or the 784 * values are invalid, set speed and duplex to -1, 785 * and return. Return 1 if speed or duplex settings are 786 * UNKNOWN; 0 otherwise. 787 */ 788 static int bond_update_speed_duplex(struct slave *slave) 789 { 790 struct net_device *slave_dev = slave->dev; 791 struct ethtool_link_ksettings ecmd; 792 int res; 793 794 slave->speed = SPEED_UNKNOWN; 795 slave->duplex = DUPLEX_UNKNOWN; 796 797 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 798 if (res < 0) 799 return 1; 800 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 801 return 1; 802 switch (ecmd.base.duplex) { 803 case DUPLEX_FULL: 804 case DUPLEX_HALF: 805 break; 806 default: 807 return 1; 808 } 809 810 slave->speed = ecmd.base.speed; 811 slave->duplex = ecmd.base.duplex; 812 813 return 0; 814 } 815 816 const char *bond_slave_link_status(s8 link) 817 { 818 switch (link) { 819 case BOND_LINK_UP: 820 return "up"; 821 case BOND_LINK_FAIL: 822 return "going down"; 823 case BOND_LINK_DOWN: 824 return "down"; 825 case BOND_LINK_BACK: 826 return "going back"; 827 default: 828 return "unknown"; 829 } 830 } 831 832 /*----------------------------- Multicast list ------------------------------*/ 833 834 /* Push the promiscuity flag down to appropriate slaves */ 835 static int bond_set_promiscuity(struct bonding *bond, int inc) 836 { 837 struct list_head *iter; 838 int err = 0; 839 840 if (bond_uses_primary(bond)) { 841 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 842 843 if (curr_active) 844 err = dev_set_promiscuity(curr_active->dev, inc); 845 } else { 846 struct slave *slave; 847 848 bond_for_each_slave(bond, slave, iter) { 849 err = dev_set_promiscuity(slave->dev, inc); 850 if (err) 851 return err; 852 } 853 } 854 return err; 855 } 856 857 /* Push the allmulti flag down to all slaves */ 858 static int bond_set_allmulti(struct bonding *bond, int inc) 859 { 860 struct list_head *iter; 861 int err = 0; 862 863 if (bond_uses_primary(bond)) { 864 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 865 866 if (curr_active) 867 err = dev_set_allmulti(curr_active->dev, inc); 868 } else { 869 struct slave *slave; 870 871 bond_for_each_slave(bond, slave, iter) { 872 err = dev_set_allmulti(slave->dev, inc); 873 if (err) 874 return err; 875 } 876 } 877 return err; 878 } 879 880 /* Retrieve the list of registered multicast addresses for the bonding 881 * device and retransmit an IGMP JOIN request to the current active 882 * slave. 883 */ 884 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 885 { 886 struct bonding *bond = container_of(work, struct bonding, 887 mcast_work.work); 888 889 if (!rtnl_trylock()) { 890 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 891 return; 892 } 893 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 894 895 if (bond->igmp_retrans > 1) { 896 bond->igmp_retrans--; 897 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 898 } 899 rtnl_unlock(); 900 } 901 902 /* Flush bond's hardware addresses from slave */ 903 static void bond_hw_addr_flush(struct net_device *bond_dev, 904 struct net_device *slave_dev) 905 { 906 struct bonding *bond = netdev_priv(bond_dev); 907 908 dev_uc_unsync(slave_dev, bond_dev); 909 dev_mc_unsync(slave_dev, bond_dev); 910 911 if (BOND_MODE(bond) == BOND_MODE_8023AD) 912 dev_mc_del(slave_dev, lacpdu_mcast_addr); 913 } 914 915 /*--------------------------- Active slave change ---------------------------*/ 916 917 /* Update the hardware address list and promisc/allmulti for the new and 918 * old active slaves (if any). Modes that are not using primary keep all 919 * slaves up date at all times; only the modes that use primary need to call 920 * this function to swap these settings during a failover. 921 */ 922 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 923 struct slave *old_active) 924 { 925 if (old_active) { 926 if (bond->dev->flags & IFF_PROMISC) 927 dev_set_promiscuity(old_active->dev, -1); 928 929 if (bond->dev->flags & IFF_ALLMULTI) 930 dev_set_allmulti(old_active->dev, -1); 931 932 if (bond->dev->flags & IFF_UP) 933 bond_hw_addr_flush(bond->dev, old_active->dev); 934 935 bond_slave_ns_maddrs_add(bond, old_active); 936 } 937 938 if (new_active) { 939 /* FIXME: Signal errors upstream. */ 940 if (bond->dev->flags & IFF_PROMISC) 941 dev_set_promiscuity(new_active->dev, 1); 942 943 if (bond->dev->flags & IFF_ALLMULTI) 944 dev_set_allmulti(new_active->dev, 1); 945 946 if (bond->dev->flags & IFF_UP) { 947 netif_addr_lock_bh(bond->dev); 948 dev_uc_sync(new_active->dev, bond->dev); 949 dev_mc_sync(new_active->dev, bond->dev); 950 netif_addr_unlock_bh(bond->dev); 951 } 952 953 bond_slave_ns_maddrs_del(bond, new_active); 954 } 955 } 956 957 /** 958 * bond_set_dev_addr - clone slave's address to bond 959 * @bond_dev: bond net device 960 * @slave_dev: slave net device 961 * 962 * Should be called with RTNL held. 963 */ 964 static int bond_set_dev_addr(struct net_device *bond_dev, 965 struct net_device *slave_dev) 966 { 967 int err; 968 969 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 970 bond_dev, slave_dev, slave_dev->addr_len); 971 err = netif_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 972 if (err) 973 return err; 974 975 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 976 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 977 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 978 return 0; 979 } 980 981 static struct slave *bond_get_old_active(struct bonding *bond, 982 struct slave *new_active) 983 { 984 struct slave *slave; 985 struct list_head *iter; 986 987 bond_for_each_slave(bond, slave, iter) { 988 if (slave == new_active) 989 continue; 990 991 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 992 return slave; 993 } 994 995 return NULL; 996 } 997 998 /* bond_do_fail_over_mac 999 * 1000 * Perform special MAC address swapping for fail_over_mac settings 1001 * 1002 * Called with RTNL 1003 */ 1004 static void bond_do_fail_over_mac(struct bonding *bond, 1005 struct slave *new_active, 1006 struct slave *old_active) 1007 { 1008 u8 tmp_mac[MAX_ADDR_LEN]; 1009 struct sockaddr_storage ss; 1010 int rv; 1011 1012 switch (bond->params.fail_over_mac) { 1013 case BOND_FOM_ACTIVE: 1014 if (new_active) { 1015 rv = bond_set_dev_addr(bond->dev, new_active->dev); 1016 if (rv) 1017 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 1018 -rv); 1019 } 1020 break; 1021 case BOND_FOM_FOLLOW: 1022 /* if new_active && old_active, swap them 1023 * if just old_active, do nothing (going to no active slave) 1024 * if just new_active, set new_active to bond's MAC 1025 */ 1026 if (!new_active) 1027 return; 1028 1029 if (!old_active) 1030 old_active = bond_get_old_active(bond, new_active); 1031 1032 if (old_active) { 1033 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 1034 new_active->dev->addr_len); 1035 bond_hw_addr_copy(ss.__data, 1036 old_active->dev->dev_addr, 1037 old_active->dev->addr_len); 1038 ss.ss_family = new_active->dev->type; 1039 } else { 1040 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 1041 bond->dev->addr_len); 1042 ss.ss_family = bond->dev->type; 1043 } 1044 1045 rv = dev_set_mac_address(new_active->dev, &ss, NULL); 1046 if (rv) { 1047 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1048 -rv); 1049 goto out; 1050 } 1051 1052 if (!old_active) 1053 goto out; 1054 1055 bond_hw_addr_copy(ss.__data, tmp_mac, 1056 new_active->dev->addr_len); 1057 ss.ss_family = old_active->dev->type; 1058 1059 rv = dev_set_mac_address(old_active->dev, &ss, NULL); 1060 if (rv) 1061 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1062 -rv); 1063 out: 1064 break; 1065 default: 1066 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1067 bond->params.fail_over_mac); 1068 break; 1069 } 1070 1071 } 1072 1073 /** 1074 * bond_choose_primary_or_current - select the primary or high priority slave 1075 * @bond: our bonding struct 1076 * 1077 * - Check if there is a primary link. If the primary link was set and is up, 1078 * go on and do link reselection. 1079 * 1080 * - If primary link is not set or down, find the highest priority link. 1081 * If the highest priority link is not current slave, set it as primary 1082 * link and do link reselection. 1083 */ 1084 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1085 { 1086 struct slave *prim = rtnl_dereference(bond->primary_slave); 1087 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1088 struct slave *slave, *hprio = NULL; 1089 struct list_head *iter; 1090 1091 if (!prim || prim->link != BOND_LINK_UP) { 1092 bond_for_each_slave(bond, slave, iter) { 1093 if (slave->link == BOND_LINK_UP) { 1094 hprio = hprio ?: slave; 1095 if (slave->prio > hprio->prio) 1096 hprio = slave; 1097 } 1098 } 1099 1100 if (hprio && hprio != curr) { 1101 prim = hprio; 1102 goto link_reselect; 1103 } 1104 1105 if (!curr || curr->link != BOND_LINK_UP) 1106 return NULL; 1107 return curr; 1108 } 1109 1110 if (bond->force_primary) { 1111 bond->force_primary = false; 1112 return prim; 1113 } 1114 1115 link_reselect: 1116 if (!curr || curr->link != BOND_LINK_UP) 1117 return prim; 1118 1119 /* At this point, prim and curr are both up */ 1120 switch (bond->params.primary_reselect) { 1121 case BOND_PRI_RESELECT_ALWAYS: 1122 return prim; 1123 case BOND_PRI_RESELECT_BETTER: 1124 if (prim->speed < curr->speed) 1125 return curr; 1126 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1127 return curr; 1128 return prim; 1129 case BOND_PRI_RESELECT_FAILURE: 1130 return curr; 1131 default: 1132 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1133 bond->params.primary_reselect); 1134 return curr; 1135 } 1136 } 1137 1138 /** 1139 * bond_find_best_slave - select the best available slave to be the active one 1140 * @bond: our bonding struct 1141 */ 1142 static struct slave *bond_find_best_slave(struct bonding *bond) 1143 { 1144 struct slave *slave, *bestslave = NULL; 1145 struct list_head *iter; 1146 int mintime = bond->params.updelay; 1147 1148 slave = bond_choose_primary_or_current(bond); 1149 if (slave) 1150 return slave; 1151 1152 bond_for_each_slave(bond, slave, iter) { 1153 if (slave->link == BOND_LINK_UP) 1154 return slave; 1155 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1156 slave->delay < mintime) { 1157 mintime = slave->delay; 1158 bestslave = slave; 1159 } 1160 } 1161 1162 return bestslave; 1163 } 1164 1165 /* must be called in RCU critical section or with RTNL held */ 1166 static bool bond_should_notify_peers(struct bonding *bond) 1167 { 1168 struct bond_up_slave *usable; 1169 struct slave *slave = NULL; 1170 1171 if (!bond->send_peer_notif || 1172 bond->send_peer_notif % 1173 max(1, bond->params.peer_notif_delay) != 0 || 1174 !netif_carrier_ok(bond->dev)) 1175 return false; 1176 1177 /* The send_peer_notif is set by active-backup or 8023ad 1178 * mode, and cleared in bond_close() when changing mode. 1179 * It is safe to only check bond mode here. 1180 */ 1181 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1182 usable = rcu_dereference_rtnl(bond->usable_slaves); 1183 if (!usable || !READ_ONCE(usable->count)) 1184 return false; 1185 } else { 1186 slave = rcu_dereference_rtnl(bond->curr_active_slave); 1187 if (!slave || test_bit(__LINK_STATE_LINKWATCH_PENDING, 1188 &slave->dev->state)) 1189 return false; 1190 } 1191 1192 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1193 slave ? slave->dev->name : "all"); 1194 1195 return true; 1196 } 1197 1198 /** 1199 * bond_change_active_slave - change the active slave into the specified one 1200 * @bond: our bonding struct 1201 * @new_active: the new slave to make the active one 1202 * 1203 * Set the new slave to the bond's settings and unset them on the old 1204 * curr_active_slave. 1205 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1206 * 1207 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1208 * because it is apparently the best available slave we have, even though its 1209 * updelay hasn't timed out yet. 1210 * 1211 * Caller must hold RTNL. 1212 */ 1213 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1214 { 1215 struct slave *old_active; 1216 1217 ASSERT_RTNL(); 1218 1219 old_active = rtnl_dereference(bond->curr_active_slave); 1220 1221 if (old_active == new_active) 1222 return; 1223 1224 #ifdef CONFIG_XFRM_OFFLOAD 1225 bond_ipsec_del_sa_all(bond); 1226 #endif /* CONFIG_XFRM_OFFLOAD */ 1227 1228 if (new_active) { 1229 new_active->last_link_up = jiffies; 1230 1231 if (new_active->link == BOND_LINK_BACK) { 1232 if (bond_uses_primary(bond)) { 1233 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1234 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1235 } 1236 1237 new_active->delay = 0; 1238 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1239 BOND_SLAVE_NOTIFY_NOW); 1240 1241 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1242 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1243 1244 if (bond_is_lb(bond)) 1245 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1246 } else { 1247 if (bond_uses_primary(bond)) 1248 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1249 } 1250 } 1251 1252 if (bond_uses_primary(bond)) 1253 bond_hw_addr_swap(bond, new_active, old_active); 1254 1255 if (bond_is_lb(bond)) { 1256 bond_alb_handle_active_change(bond, new_active); 1257 if (old_active) 1258 bond_set_slave_inactive_flags(old_active, 1259 BOND_SLAVE_NOTIFY_NOW); 1260 if (new_active) 1261 bond_set_slave_active_flags(new_active, 1262 BOND_SLAVE_NOTIFY_NOW); 1263 } else { 1264 rcu_assign_pointer(bond->curr_active_slave, new_active); 1265 } 1266 1267 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1268 if (old_active) 1269 bond_set_slave_inactive_flags(old_active, 1270 BOND_SLAVE_NOTIFY_NOW); 1271 1272 if (new_active) { 1273 bool should_notify_peers = false; 1274 1275 bond_set_slave_active_flags(new_active, 1276 BOND_SLAVE_NOTIFY_NOW); 1277 1278 if (bond->params.fail_over_mac) 1279 bond_do_fail_over_mac(bond, new_active, 1280 old_active); 1281 1282 if (netif_running(bond->dev)) { 1283 bond->send_peer_notif = 1284 bond->params.num_peer_notif * 1285 max(1, bond->params.peer_notif_delay); 1286 should_notify_peers = 1287 bond_should_notify_peers(bond); 1288 } 1289 1290 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1291 if (should_notify_peers) { 1292 bond->send_peer_notif--; 1293 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1294 bond->dev); 1295 } 1296 } 1297 } 1298 1299 #ifdef CONFIG_XFRM_OFFLOAD 1300 bond_ipsec_add_sa_all(bond); 1301 #endif /* CONFIG_XFRM_OFFLOAD */ 1302 1303 /* resend IGMP joins since active slave has changed or 1304 * all were sent on curr_active_slave. 1305 * resend only if bond is brought up with the affected 1306 * bonding modes and the retransmission is enabled 1307 */ 1308 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1309 ((bond_uses_primary(bond) && new_active) || 1310 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1311 bond->igmp_retrans = bond->params.resend_igmp; 1312 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1313 } 1314 } 1315 1316 /** 1317 * bond_select_active_slave - select a new active slave, if needed 1318 * @bond: our bonding struct 1319 * 1320 * This functions should be called when one of the following occurs: 1321 * - The old curr_active_slave has been released or lost its link. 1322 * - The primary_slave has got its link back. 1323 * - A slave has got its link back and there's no old curr_active_slave. 1324 * 1325 * Caller must hold RTNL. 1326 */ 1327 void bond_select_active_slave(struct bonding *bond) 1328 { 1329 struct slave *best_slave; 1330 int rv; 1331 1332 ASSERT_RTNL(); 1333 1334 best_slave = bond_find_best_slave(bond); 1335 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1336 bond_change_active_slave(bond, best_slave); 1337 rv = bond_set_carrier(bond); 1338 if (!rv) 1339 return; 1340 1341 if (netif_carrier_ok(bond->dev)) 1342 netdev_info(bond->dev, "active interface up!\n"); 1343 else 1344 netdev_info(bond->dev, "now running without any active interface!\n"); 1345 } 1346 } 1347 1348 #ifdef CONFIG_NET_POLL_CONTROLLER 1349 static inline int slave_enable_netpoll(struct slave *slave) 1350 { 1351 struct netpoll *np; 1352 int err = 0; 1353 1354 np = kzalloc(sizeof(*np), GFP_KERNEL); 1355 err = -ENOMEM; 1356 if (!np) 1357 goto out; 1358 1359 err = __netpoll_setup(np, slave->dev); 1360 if (err) { 1361 kfree(np); 1362 goto out; 1363 } 1364 slave->np = np; 1365 out: 1366 return err; 1367 } 1368 static inline void slave_disable_netpoll(struct slave *slave) 1369 { 1370 struct netpoll *np = slave->np; 1371 1372 if (!np) 1373 return; 1374 1375 slave->np = NULL; 1376 1377 __netpoll_free(np); 1378 } 1379 1380 static void bond_poll_controller(struct net_device *bond_dev) 1381 { 1382 struct bonding *bond = netdev_priv(bond_dev); 1383 struct slave *slave = NULL; 1384 struct list_head *iter; 1385 struct ad_info ad_info; 1386 1387 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1388 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1389 return; 1390 1391 bond_for_each_slave_rcu(bond, slave, iter) { 1392 if (!bond_slave_is_up(slave)) 1393 continue; 1394 1395 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1396 struct aggregator *agg = 1397 SLAVE_AD_INFO(slave)->port.aggregator; 1398 1399 if (agg && 1400 agg->aggregator_identifier != ad_info.aggregator_id) 1401 continue; 1402 } 1403 1404 netpoll_poll_dev(slave->dev); 1405 } 1406 } 1407 1408 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1409 { 1410 struct bonding *bond = netdev_priv(bond_dev); 1411 struct list_head *iter; 1412 struct slave *slave; 1413 1414 bond_for_each_slave(bond, slave, iter) 1415 if (bond_slave_is_up(slave)) 1416 slave_disable_netpoll(slave); 1417 } 1418 1419 static int bond_netpoll_setup(struct net_device *dev) 1420 { 1421 struct bonding *bond = netdev_priv(dev); 1422 struct list_head *iter; 1423 struct slave *slave; 1424 int err = 0; 1425 1426 bond_for_each_slave(bond, slave, iter) { 1427 err = slave_enable_netpoll(slave); 1428 if (err) { 1429 bond_netpoll_cleanup(dev); 1430 break; 1431 } 1432 } 1433 return err; 1434 } 1435 #else 1436 static inline int slave_enable_netpoll(struct slave *slave) 1437 { 1438 return 0; 1439 } 1440 static inline void slave_disable_netpoll(struct slave *slave) 1441 { 1442 } 1443 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1444 { 1445 } 1446 #endif 1447 1448 /*---------------------------------- IOCTL ----------------------------------*/ 1449 1450 static netdev_features_t bond_fix_features(struct net_device *dev, 1451 netdev_features_t features) 1452 { 1453 struct bonding *bond = netdev_priv(dev); 1454 struct list_head *iter; 1455 netdev_features_t mask; 1456 struct slave *slave; 1457 1458 mask = features; 1459 features = netdev_base_features(features); 1460 1461 bond_for_each_slave(bond, slave, iter) { 1462 features = netdev_increment_features(features, 1463 slave->dev->features, 1464 mask); 1465 } 1466 features = netdev_add_tso_features(features, mask); 1467 1468 return features; 1469 } 1470 1471 static void bond_setup_by_slave(struct net_device *bond_dev, 1472 struct net_device *slave_dev) 1473 { 1474 bool was_up = !!(bond_dev->flags & IFF_UP); 1475 1476 dev_close(bond_dev); 1477 1478 bond_dev->header_ops = slave_dev->header_ops; 1479 1480 bond_dev->type = slave_dev->type; 1481 bond_dev->hard_header_len = slave_dev->hard_header_len; 1482 bond_dev->needed_headroom = slave_dev->needed_headroom; 1483 bond_dev->addr_len = slave_dev->addr_len; 1484 1485 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1486 slave_dev->addr_len); 1487 1488 if (slave_dev->flags & IFF_POINTOPOINT) { 1489 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST); 1490 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP); 1491 } 1492 if (was_up) 1493 dev_open(bond_dev, NULL); 1494 } 1495 1496 /* On bonding slaves other than the currently active slave, suppress 1497 * duplicates except for alb non-mcast/bcast. 1498 */ 1499 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1500 struct slave *slave, 1501 struct bonding *bond) 1502 { 1503 if (bond_is_slave_inactive(slave)) { 1504 if (BOND_MODE(bond) == BOND_MODE_ALB && 1505 skb->pkt_type != PACKET_BROADCAST && 1506 skb->pkt_type != PACKET_MULTICAST) 1507 return false; 1508 return true; 1509 } 1510 return false; 1511 } 1512 1513 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1514 { 1515 struct sk_buff *skb = *pskb; 1516 struct slave *slave; 1517 struct bonding *bond; 1518 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1519 struct slave *); 1520 int ret = RX_HANDLER_ANOTHER; 1521 1522 skb = skb_share_check(skb, GFP_ATOMIC); 1523 if (unlikely(!skb)) 1524 return RX_HANDLER_CONSUMED; 1525 1526 *pskb = skb; 1527 1528 slave = bond_slave_get_rcu(skb->dev); 1529 bond = slave->bond; 1530 1531 recv_probe = READ_ONCE(bond->recv_probe); 1532 if (recv_probe) { 1533 ret = recv_probe(skb, bond, slave); 1534 if (ret == RX_HANDLER_CONSUMED) { 1535 consume_skb(skb); 1536 return ret; 1537 } 1538 } 1539 1540 /* 1541 * For packets determined by bond_should_deliver_exact_match() call to 1542 * be suppressed we want to make an exception for link-local packets. 1543 * This is necessary for e.g. LLDP daemons to be able to monitor 1544 * inactive slave links without being forced to bind to them 1545 * explicitly. 1546 * 1547 * At the same time, packets that are passed to the bonding master 1548 * (including link-local ones) can have their originating interface 1549 * determined via PACKET_ORIGDEV socket option. 1550 */ 1551 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1552 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1553 return RX_HANDLER_PASS; 1554 return RX_HANDLER_EXACT; 1555 } 1556 1557 skb->dev = bond->dev; 1558 1559 if (BOND_MODE(bond) == BOND_MODE_ALB && 1560 netif_is_bridge_port(bond->dev) && 1561 skb->pkt_type == PACKET_HOST) { 1562 1563 if (unlikely(skb_cow_head(skb, 1564 skb->data - skb_mac_header(skb)))) { 1565 kfree_skb(skb); 1566 return RX_HANDLER_CONSUMED; 1567 } 1568 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1569 bond->dev->addr_len); 1570 } 1571 1572 return ret; 1573 } 1574 1575 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1576 { 1577 switch (BOND_MODE(bond)) { 1578 case BOND_MODE_ROUNDROBIN: 1579 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1580 case BOND_MODE_ACTIVEBACKUP: 1581 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1582 case BOND_MODE_BROADCAST: 1583 return NETDEV_LAG_TX_TYPE_BROADCAST; 1584 case BOND_MODE_XOR: 1585 case BOND_MODE_8023AD: 1586 return NETDEV_LAG_TX_TYPE_HASH; 1587 default: 1588 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1589 } 1590 } 1591 1592 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1593 enum netdev_lag_tx_type type) 1594 { 1595 if (type != NETDEV_LAG_TX_TYPE_HASH) 1596 return NETDEV_LAG_HASH_NONE; 1597 1598 switch (bond->params.xmit_policy) { 1599 case BOND_XMIT_POLICY_LAYER2: 1600 return NETDEV_LAG_HASH_L2; 1601 case BOND_XMIT_POLICY_LAYER34: 1602 return NETDEV_LAG_HASH_L34; 1603 case BOND_XMIT_POLICY_LAYER23: 1604 return NETDEV_LAG_HASH_L23; 1605 case BOND_XMIT_POLICY_ENCAP23: 1606 return NETDEV_LAG_HASH_E23; 1607 case BOND_XMIT_POLICY_ENCAP34: 1608 return NETDEV_LAG_HASH_E34; 1609 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1610 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1611 default: 1612 return NETDEV_LAG_HASH_UNKNOWN; 1613 } 1614 } 1615 1616 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1617 struct netlink_ext_ack *extack) 1618 { 1619 struct netdev_lag_upper_info lag_upper_info; 1620 enum netdev_lag_tx_type type; 1621 int err; 1622 1623 type = bond_lag_tx_type(bond); 1624 lag_upper_info.tx_type = type; 1625 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1626 1627 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1628 &lag_upper_info, extack); 1629 if (err) 1630 return err; 1631 1632 slave->dev->flags |= IFF_SLAVE; 1633 return 0; 1634 } 1635 1636 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1637 { 1638 netdev_upper_dev_unlink(slave->dev, bond->dev); 1639 slave->dev->flags &= ~IFF_SLAVE; 1640 } 1641 1642 static void slave_kobj_release(struct kobject *kobj) 1643 { 1644 struct slave *slave = to_slave(kobj); 1645 struct bonding *bond = bond_get_bond_by_slave(slave); 1646 1647 cancel_delayed_work_sync(&slave->notify_work); 1648 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1649 kfree(SLAVE_AD_INFO(slave)); 1650 1651 kfree(slave); 1652 } 1653 1654 static struct kobj_type slave_ktype = { 1655 .release = slave_kobj_release, 1656 #ifdef CONFIG_SYSFS 1657 .sysfs_ops = &slave_sysfs_ops, 1658 #endif 1659 }; 1660 1661 static int bond_kobj_init(struct slave *slave) 1662 { 1663 int err; 1664 1665 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1666 &(slave->dev->dev.kobj), "bonding_slave"); 1667 if (err) 1668 kobject_put(&slave->kobj); 1669 1670 return err; 1671 } 1672 1673 static struct slave *bond_alloc_slave(struct bonding *bond, 1674 struct net_device *slave_dev) 1675 { 1676 struct slave *slave = NULL; 1677 1678 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1679 if (!slave) 1680 return NULL; 1681 1682 slave->bond = bond; 1683 slave->dev = slave_dev; 1684 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1685 1686 if (bond_kobj_init(slave)) 1687 return NULL; 1688 1689 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1690 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1691 GFP_KERNEL); 1692 if (!SLAVE_AD_INFO(slave)) { 1693 kobject_put(&slave->kobj); 1694 return NULL; 1695 } 1696 } 1697 1698 return slave; 1699 } 1700 1701 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1702 { 1703 info->bond_mode = BOND_MODE(bond); 1704 info->miimon = bond->params.miimon; 1705 info->num_slaves = bond->slave_cnt; 1706 } 1707 1708 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1709 { 1710 strcpy(info->slave_name, slave->dev->name); 1711 info->link = slave->link; 1712 info->state = bond_slave_state(slave); 1713 info->link_failure_count = slave->link_failure_count; 1714 } 1715 1716 static void bond_netdev_notify_work(struct work_struct *_work) 1717 { 1718 struct slave *slave = container_of(_work, struct slave, 1719 notify_work.work); 1720 1721 if (rtnl_trylock()) { 1722 struct netdev_bonding_info binfo; 1723 1724 bond_fill_ifslave(slave, &binfo.slave); 1725 bond_fill_ifbond(slave->bond, &binfo.master); 1726 netdev_bonding_info_change(slave->dev, &binfo); 1727 rtnl_unlock(); 1728 } else { 1729 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1730 } 1731 } 1732 1733 void bond_queue_slave_event(struct slave *slave) 1734 { 1735 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1736 } 1737 1738 void bond_lower_state_changed(struct slave *slave) 1739 { 1740 struct netdev_lag_lower_state_info info; 1741 1742 info.link_up = slave->link == BOND_LINK_UP || 1743 slave->link == BOND_LINK_FAIL; 1744 info.tx_enabled = bond_is_active_slave(slave); 1745 netdev_lower_state_changed(slave->dev, &info); 1746 } 1747 1748 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1749 if (extack) \ 1750 NL_SET_ERR_MSG(extack, errmsg); \ 1751 else \ 1752 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1753 } while (0) 1754 1755 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1756 if (extack) \ 1757 NL_SET_ERR_MSG(extack, errmsg); \ 1758 else \ 1759 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1760 } while (0) 1761 1762 /* The bonding driver uses ether_setup() to convert a master bond device 1763 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always 1764 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP 1765 * if they were set 1766 */ 1767 static void bond_ether_setup(struct net_device *bond_dev) 1768 { 1769 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP); 1770 1771 ether_setup(bond_dev); 1772 bond_dev->flags |= IFF_MASTER | flags; 1773 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1774 } 1775 1776 void bond_xdp_set_features(struct net_device *bond_dev) 1777 { 1778 struct bonding *bond = netdev_priv(bond_dev); 1779 xdp_features_t val = NETDEV_XDP_ACT_MASK; 1780 struct list_head *iter; 1781 struct slave *slave; 1782 1783 ASSERT_RTNL(); 1784 1785 if (!bond_xdp_check(bond, BOND_MODE(bond)) || !bond_has_slaves(bond)) { 1786 xdp_clear_features_flag(bond_dev); 1787 return; 1788 } 1789 1790 bond_for_each_slave(bond, slave, iter) 1791 val &= slave->dev->xdp_features; 1792 1793 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY; 1794 1795 xdp_set_features_flag(bond_dev, val); 1796 } 1797 1798 /* enslave device <slave> to bond device <master> */ 1799 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1800 struct netlink_ext_ack *extack) 1801 { 1802 struct bonding *bond = netdev_priv(bond_dev); 1803 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1804 struct slave *new_slave = NULL, *prev_slave; 1805 struct sockaddr_storage ss; 1806 int res = 0, i; 1807 1808 if (slave_dev->flags & IFF_MASTER && 1809 !netif_is_bond_master(slave_dev)) { 1810 BOND_NL_ERR(bond_dev, extack, 1811 "Device type (master device) cannot be enslaved"); 1812 return -EPERM; 1813 } 1814 1815 /* already in-use? */ 1816 if (netdev_is_rx_handler_busy(slave_dev)) { 1817 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1818 "Device is in use and cannot be enslaved"); 1819 return -EBUSY; 1820 } 1821 1822 if (bond_dev == slave_dev) { 1823 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1824 return -EPERM; 1825 } 1826 1827 /* vlan challenged mutual exclusion */ 1828 /* no need to lock since we're protected by rtnl_lock */ 1829 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1830 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1831 if (vlan_uses_dev(bond_dev)) { 1832 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1833 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1834 return -EPERM; 1835 } else { 1836 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1837 } 1838 } else { 1839 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1840 } 1841 1842 if (slave_dev->features & NETIF_F_HW_ESP) 1843 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1844 1845 /* Old ifenslave binaries are no longer supported. These can 1846 * be identified with moderate accuracy by the state of the slave: 1847 * the current ifenslave will set the interface down prior to 1848 * enslaving it; the old ifenslave will not. 1849 */ 1850 if (slave_dev->flags & IFF_UP) { 1851 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1852 "Device can not be enslaved while up"); 1853 return -EPERM; 1854 } 1855 1856 /* set bonding device ether type by slave - bonding netdevices are 1857 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1858 * there is a need to override some of the type dependent attribs/funcs. 1859 * 1860 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1861 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1862 */ 1863 if (!bond_has_slaves(bond)) { 1864 if (bond_dev->type != slave_dev->type) { 1865 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1866 bond_dev->type, slave_dev->type); 1867 1868 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1869 bond_dev); 1870 res = notifier_to_errno(res); 1871 if (res) { 1872 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1873 return -EBUSY; 1874 } 1875 1876 /* Flush unicast and multicast addresses */ 1877 dev_uc_flush(bond_dev); 1878 dev_mc_flush(bond_dev); 1879 1880 if (slave_dev->type != ARPHRD_ETHER) 1881 bond_setup_by_slave(bond_dev, slave_dev); 1882 else 1883 bond_ether_setup(bond_dev); 1884 1885 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1886 bond_dev); 1887 } 1888 } else if (bond_dev->type != slave_dev->type) { 1889 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1890 "Device type is different from other slaves"); 1891 return -EINVAL; 1892 } 1893 1894 if (slave_dev->type == ARPHRD_INFINIBAND && 1895 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1896 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1897 "Only active-backup mode is supported for infiniband slaves"); 1898 res = -EOPNOTSUPP; 1899 goto err_undo_flags; 1900 } 1901 1902 if (!slave_ops->ndo_set_mac_address || 1903 slave_dev->type == ARPHRD_INFINIBAND) { 1904 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1905 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1906 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1907 if (!bond_has_slaves(bond)) { 1908 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1909 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1910 } else { 1911 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1912 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1913 res = -EOPNOTSUPP; 1914 goto err_undo_flags; 1915 } 1916 } 1917 } 1918 1919 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1920 1921 /* If this is the first slave, then we need to set the master's hardware 1922 * address to be the same as the slave's. 1923 */ 1924 if (!bond_has_slaves(bond) && 1925 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1926 res = bond_set_dev_addr(bond->dev, slave_dev); 1927 if (res) 1928 goto err_undo_flags; 1929 } 1930 1931 new_slave = bond_alloc_slave(bond, slave_dev); 1932 if (!new_slave) { 1933 res = -ENOMEM; 1934 goto err_undo_flags; 1935 } 1936 1937 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1938 * is set via sysfs or module option if desired. 1939 */ 1940 new_slave->queue_id = 0; 1941 1942 /* Save slave's original mtu and then set it to match the bond */ 1943 new_slave->original_mtu = slave_dev->mtu; 1944 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1945 if (res) { 1946 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1947 goto err_free; 1948 } 1949 1950 /* Save slave's original ("permanent") mac address for modes 1951 * that need it, and for restoring it upon release, and then 1952 * set it to the master's address 1953 */ 1954 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1955 slave_dev->addr_len); 1956 1957 if (!bond->params.fail_over_mac || 1958 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1959 /* Set slave to master's mac address. The application already 1960 * set the master's mac address to that of the first slave 1961 */ 1962 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1963 } else if (bond->params.fail_over_mac == BOND_FOM_FOLLOW && 1964 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1965 bond_has_slaves(bond) && 1966 memcmp(slave_dev->dev_addr, bond_dev->dev_addr, bond_dev->addr_len) == 0) { 1967 /* Set slave to random address to avoid duplicate mac 1968 * address in later fail over. 1969 */ 1970 eth_random_addr(ss.__data); 1971 } else { 1972 goto skip_mac_set; 1973 } 1974 1975 ss.ss_family = slave_dev->type; 1976 res = dev_set_mac_address(slave_dev, &ss, extack); 1977 if (res) { 1978 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1979 goto err_restore_mtu; 1980 } 1981 1982 skip_mac_set: 1983 1984 /* set no_addrconf flag before open to prevent IPv6 addrconf */ 1985 slave_dev->priv_flags |= IFF_NO_ADDRCONF; 1986 1987 /* open the slave since the application closed it */ 1988 res = dev_open(slave_dev, extack); 1989 if (res) { 1990 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1991 goto err_restore_mac; 1992 } 1993 1994 slave_dev->priv_flags |= IFF_BONDING; 1995 /* initialize slave stats */ 1996 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1997 1998 if (bond_is_lb(bond)) { 1999 /* bond_alb_init_slave() must be called before all other stages since 2000 * it might fail and we do not want to have to undo everything 2001 */ 2002 res = bond_alb_init_slave(bond, new_slave); 2003 if (res) 2004 goto err_close; 2005 } 2006 2007 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 2008 if (res) { 2009 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 2010 goto err_close; 2011 } 2012 2013 prev_slave = bond_last_slave(bond); 2014 2015 new_slave->delay = 0; 2016 new_slave->link_failure_count = 0; 2017 2018 if (bond_update_speed_duplex(new_slave) && 2019 bond_needs_speed_duplex(bond)) 2020 new_slave->link = BOND_LINK_DOWN; 2021 2022 new_slave->last_rx = jiffies - 2023 (msecs_to_jiffies(bond->params.arp_interval) + 1); 2024 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2025 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2026 2027 new_slave->last_tx = new_slave->last_rx; 2028 2029 /* check for initial state */ 2030 new_slave->link = BOND_LINK_NOCHANGE; 2031 if (bond->params.miimon) { 2032 if (netif_carrier_ok(slave_dev)) { 2033 if (bond->params.updelay) { 2034 bond_set_slave_link_state(new_slave, 2035 BOND_LINK_BACK, 2036 BOND_SLAVE_NOTIFY_NOW); 2037 new_slave->delay = bond->params.updelay; 2038 } else { 2039 bond_set_slave_link_state(new_slave, 2040 BOND_LINK_UP, 2041 BOND_SLAVE_NOTIFY_NOW); 2042 } 2043 } else { 2044 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2045 BOND_SLAVE_NOTIFY_NOW); 2046 } 2047 } else if (bond->params.arp_interval) { 2048 bond_set_slave_link_state(new_slave, 2049 (netif_carrier_ok(slave_dev) ? 2050 BOND_LINK_UP : BOND_LINK_DOWN), 2051 BOND_SLAVE_NOTIFY_NOW); 2052 } else { 2053 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2054 BOND_SLAVE_NOTIFY_NOW); 2055 } 2056 2057 if (new_slave->link != BOND_LINK_DOWN) 2058 new_slave->last_link_up = jiffies; 2059 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2060 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2061 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2062 2063 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2064 /* if there is a primary slave, remember it */ 2065 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2066 rcu_assign_pointer(bond->primary_slave, new_slave); 2067 bond->force_primary = true; 2068 } 2069 } 2070 2071 switch (BOND_MODE(bond)) { 2072 case BOND_MODE_ACTIVEBACKUP: 2073 bond_set_slave_inactive_flags(new_slave, 2074 BOND_SLAVE_NOTIFY_NOW); 2075 break; 2076 case BOND_MODE_8023AD: 2077 /* in 802.3ad mode, the internal mechanism 2078 * will activate the slaves in the selected 2079 * aggregator 2080 */ 2081 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2082 /* if this is the first slave */ 2083 if (!prev_slave) { 2084 SLAVE_AD_INFO(new_slave)->id = 1; 2085 /* Initialize AD with the number of times that the AD timer is called in 1 second 2086 * can be called only after the mac address of the bond is set 2087 */ 2088 bond_3ad_initialize(bond); 2089 } else { 2090 SLAVE_AD_INFO(new_slave)->id = 2091 SLAVE_AD_INFO(prev_slave)->id + 1; 2092 } 2093 2094 bond_3ad_bind_slave(new_slave); 2095 break; 2096 case BOND_MODE_TLB: 2097 case BOND_MODE_ALB: 2098 bond_set_active_slave(new_slave); 2099 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2100 break; 2101 default: 2102 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2103 2104 /* always active in trunk mode */ 2105 bond_set_active_slave(new_slave); 2106 2107 /* In trunking mode there is little meaning to curr_active_slave 2108 * anyway (it holds no special properties of the bond device), 2109 * so we can change it without calling change_active_interface() 2110 */ 2111 if (!rcu_access_pointer(bond->curr_active_slave) && 2112 new_slave->link == BOND_LINK_UP) 2113 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2114 2115 break; 2116 } /* switch(bond_mode) */ 2117 2118 #ifdef CONFIG_NET_POLL_CONTROLLER 2119 if (bond->dev->npinfo) { 2120 if (slave_enable_netpoll(new_slave)) { 2121 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2122 res = -EBUSY; 2123 goto err_detach; 2124 } 2125 } 2126 #endif 2127 2128 if (!(bond_dev->features & NETIF_F_LRO)) 2129 dev_disable_lro(slave_dev); 2130 2131 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2132 new_slave); 2133 if (res) { 2134 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2135 goto err_detach; 2136 } 2137 2138 res = bond_master_upper_dev_link(bond, new_slave, extack); 2139 if (res) { 2140 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2141 goto err_unregister; 2142 } 2143 2144 bond_lower_state_changed(new_slave); 2145 2146 res = bond_sysfs_slave_add(new_slave); 2147 if (res) { 2148 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2149 goto err_upper_unlink; 2150 } 2151 2152 /* If the mode uses primary, then the following is handled by 2153 * bond_change_active_slave(). 2154 */ 2155 if (!bond_uses_primary(bond)) { 2156 /* set promiscuity level to new slave */ 2157 if (bond_dev->flags & IFF_PROMISC) { 2158 res = dev_set_promiscuity(slave_dev, 1); 2159 if (res) 2160 goto err_sysfs_del; 2161 } 2162 2163 /* set allmulti level to new slave */ 2164 if (bond_dev->flags & IFF_ALLMULTI) { 2165 res = dev_set_allmulti(slave_dev, 1); 2166 if (res) { 2167 if (bond_dev->flags & IFF_PROMISC) 2168 dev_set_promiscuity(slave_dev, -1); 2169 goto err_sysfs_del; 2170 } 2171 } 2172 2173 if (bond_dev->flags & IFF_UP) { 2174 netif_addr_lock_bh(bond_dev); 2175 dev_mc_sync_multiple(slave_dev, bond_dev); 2176 dev_uc_sync_multiple(slave_dev, bond_dev); 2177 netif_addr_unlock_bh(bond_dev); 2178 2179 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2180 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2181 } 2182 } 2183 2184 bond->slave_cnt++; 2185 netdev_compute_master_upper_features(bond->dev, true); 2186 bond_set_carrier(bond); 2187 2188 /* Needs to be called before bond_select_active_slave(), which will 2189 * remove the maddrs if the slave is selected as active slave. 2190 */ 2191 bond_slave_ns_maddrs_add(bond, new_slave); 2192 2193 if (bond_uses_primary(bond)) { 2194 block_netpoll_tx(); 2195 bond_select_active_slave(bond); 2196 unblock_netpoll_tx(); 2197 } 2198 2199 /* broadcast mode uses the all_slaves to loop through slaves. */ 2200 if (bond_mode_can_use_xmit_hash(bond) || 2201 BOND_MODE(bond) == BOND_MODE_BROADCAST) 2202 bond_update_slave_arr(bond, NULL); 2203 2204 if (!slave_dev->netdev_ops->ndo_bpf || 2205 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2206 if (bond->xdp_prog) { 2207 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2208 "Slave does not support XDP"); 2209 res = -EOPNOTSUPP; 2210 goto err_sysfs_del; 2211 } 2212 } else if (bond->xdp_prog) { 2213 struct netdev_bpf xdp = { 2214 .command = XDP_SETUP_PROG, 2215 .flags = 0, 2216 .prog = bond->xdp_prog, 2217 .extack = extack, 2218 }; 2219 2220 if (dev_xdp_prog_count(slave_dev) > 0) { 2221 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2222 "Slave has XDP program loaded, please unload before enslaving"); 2223 res = -EOPNOTSUPP; 2224 goto err_sysfs_del; 2225 } 2226 2227 res = dev_xdp_propagate(slave_dev, &xdp); 2228 if (res < 0) { 2229 /* ndo_bpf() sets extack error message */ 2230 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2231 goto err_sysfs_del; 2232 } 2233 if (bond->xdp_prog) 2234 bpf_prog_inc(bond->xdp_prog); 2235 } 2236 2237 bond_xdp_set_features(bond_dev); 2238 2239 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2240 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2241 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2242 2243 /* enslave is successful */ 2244 bond_queue_slave_event(new_slave); 2245 return 0; 2246 2247 /* Undo stages on error */ 2248 err_sysfs_del: 2249 bond_sysfs_slave_del(new_slave); 2250 2251 err_upper_unlink: 2252 bond_upper_dev_unlink(bond, new_slave); 2253 2254 err_unregister: 2255 netdev_rx_handler_unregister(slave_dev); 2256 2257 err_detach: 2258 vlan_vids_del_by_dev(slave_dev, bond_dev); 2259 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2260 RCU_INIT_POINTER(bond->primary_slave, NULL); 2261 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2262 block_netpoll_tx(); 2263 bond_change_active_slave(bond, NULL); 2264 bond_select_active_slave(bond); 2265 unblock_netpoll_tx(); 2266 } 2267 /* either primary_slave or curr_active_slave might've changed */ 2268 synchronize_rcu(); 2269 slave_disable_netpoll(new_slave); 2270 2271 err_close: 2272 if (!netif_is_bond_master(slave_dev)) 2273 slave_dev->priv_flags &= ~IFF_BONDING; 2274 dev_close(slave_dev); 2275 2276 err_restore_mac: 2277 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2278 if (!bond->params.fail_over_mac || 2279 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2280 /* XXX TODO - fom follow mode needs to change master's 2281 * MAC if this slave's MAC is in use by the bond, or at 2282 * least print a warning. 2283 */ 2284 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2285 new_slave->dev->addr_len); 2286 ss.ss_family = slave_dev->type; 2287 dev_set_mac_address(slave_dev, &ss, NULL); 2288 } 2289 2290 err_restore_mtu: 2291 dev_set_mtu(slave_dev, new_slave->original_mtu); 2292 2293 err_free: 2294 kobject_put(&new_slave->kobj); 2295 2296 err_undo_flags: 2297 /* Enslave of first slave has failed and we need to fix master's mac */ 2298 if (!bond_has_slaves(bond)) { 2299 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2300 slave_dev->dev_addr)) 2301 eth_hw_addr_random(bond_dev); 2302 if (bond_dev->type != ARPHRD_ETHER) { 2303 dev_close(bond_dev); 2304 bond_ether_setup(bond_dev); 2305 } 2306 } 2307 2308 return res; 2309 } 2310 2311 /* Try to release the slave device <slave> from the bond device <master> 2312 * It is legal to access curr_active_slave without a lock because all the function 2313 * is RTNL-locked. If "all" is true it means that the function is being called 2314 * while destroying a bond interface and all slaves are being released. 2315 * 2316 * The rules for slave state should be: 2317 * for Active/Backup: 2318 * Active stays on all backups go down 2319 * for Bonded connections: 2320 * The first up interface should be left on and all others downed. 2321 */ 2322 static int __bond_release_one(struct net_device *bond_dev, 2323 struct net_device *slave_dev, 2324 bool all, bool unregister) 2325 { 2326 struct bonding *bond = netdev_priv(bond_dev); 2327 struct slave *slave, *oldcurrent; 2328 struct sockaddr_storage ss; 2329 int old_flags = bond_dev->flags; 2330 netdev_features_t old_features = bond_dev->features; 2331 2332 /* slave is not a slave or master is not master of this slave */ 2333 if (!(slave_dev->flags & IFF_SLAVE) || 2334 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2335 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2336 return -EINVAL; 2337 } 2338 2339 block_netpoll_tx(); 2340 2341 slave = bond_get_slave_by_dev(bond, slave_dev); 2342 if (!slave) { 2343 /* not a slave of this bond */ 2344 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2345 unblock_netpoll_tx(); 2346 return -EINVAL; 2347 } 2348 2349 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2350 2351 bond_sysfs_slave_del(slave); 2352 2353 /* recompute stats just before removing the slave */ 2354 bond_get_stats(bond->dev, &bond->bond_stats); 2355 2356 if (bond->xdp_prog) { 2357 struct netdev_bpf xdp = { 2358 .command = XDP_SETUP_PROG, 2359 .flags = 0, 2360 .prog = NULL, 2361 .extack = NULL, 2362 }; 2363 if (dev_xdp_propagate(slave_dev, &xdp)) 2364 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2365 } 2366 2367 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2368 * for this slave anymore. 2369 */ 2370 netdev_rx_handler_unregister(slave_dev); 2371 2372 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2373 bond_3ad_unbind_slave(slave); 2374 2375 bond_upper_dev_unlink(bond, slave); 2376 2377 if (bond_mode_can_use_xmit_hash(bond) || 2378 BOND_MODE(bond) == BOND_MODE_BROADCAST) 2379 bond_update_slave_arr(bond, slave); 2380 2381 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2382 bond_is_active_slave(slave) ? "active" : "backup"); 2383 2384 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2385 2386 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2387 2388 if (!all && (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2389 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2390 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2391 bond_has_slaves(bond)) 2392 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n", 2393 slave->perm_hwaddr); 2394 } 2395 2396 if (rtnl_dereference(bond->primary_slave) == slave) 2397 RCU_INIT_POINTER(bond->primary_slave, NULL); 2398 2399 if (oldcurrent == slave) 2400 bond_change_active_slave(bond, NULL); 2401 2402 /* Must be called after bond_change_active_slave () as the slave 2403 * might change from an active slave to a backup slave. Then it is 2404 * necessary to clear the maddrs on the backup slave. 2405 */ 2406 bond_slave_ns_maddrs_del(bond, slave); 2407 2408 if (bond_is_lb(bond)) { 2409 /* Must be called only after the slave has been 2410 * detached from the list and the curr_active_slave 2411 * has been cleared (if our_slave == old_current), 2412 * but before a new active slave is selected. 2413 */ 2414 bond_alb_deinit_slave(bond, slave); 2415 } 2416 2417 if (all) { 2418 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2419 } else if (oldcurrent == slave) { 2420 /* Note that we hold RTNL over this sequence, so there 2421 * is no concern that another slave add/remove event 2422 * will interfere. 2423 */ 2424 bond_select_active_slave(bond); 2425 } 2426 2427 bond_set_carrier(bond); 2428 if (!bond_has_slaves(bond)) 2429 eth_hw_addr_random(bond_dev); 2430 2431 unblock_netpoll_tx(); 2432 synchronize_rcu(); 2433 bond->slave_cnt--; 2434 2435 if (!bond_has_slaves(bond)) { 2436 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2437 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2438 } 2439 2440 netdev_compute_master_upper_features(bond->dev, true); 2441 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2442 (old_features & NETIF_F_VLAN_CHALLENGED)) 2443 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2444 2445 vlan_vids_del_by_dev(slave_dev, bond_dev); 2446 2447 /* If the mode uses primary, then this case was handled above by 2448 * bond_change_active_slave(..., NULL) 2449 */ 2450 if (!bond_uses_primary(bond)) { 2451 /* unset promiscuity level from slave 2452 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2453 * of the IFF_PROMISC flag in the bond_dev, but we need the 2454 * value of that flag before that change, as that was the value 2455 * when this slave was attached, so we cache at the start of the 2456 * function and use it here. Same goes for ALLMULTI below 2457 */ 2458 if (old_flags & IFF_PROMISC) 2459 dev_set_promiscuity(slave_dev, -1); 2460 2461 /* unset allmulti level from slave */ 2462 if (old_flags & IFF_ALLMULTI) 2463 dev_set_allmulti(slave_dev, -1); 2464 2465 if (old_flags & IFF_UP) 2466 bond_hw_addr_flush(bond_dev, slave_dev); 2467 } 2468 2469 slave_disable_netpoll(slave); 2470 2471 /* close slave before restoring its mac address */ 2472 dev_close(slave_dev); 2473 2474 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2475 2476 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2477 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2478 /* restore original ("permanent") mac address */ 2479 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2480 slave->dev->addr_len); 2481 ss.ss_family = slave_dev->type; 2482 dev_set_mac_address(slave_dev, &ss, NULL); 2483 } 2484 2485 if (unregister) { 2486 netdev_lock_ops(slave_dev); 2487 __netif_set_mtu(slave_dev, slave->original_mtu); 2488 netdev_unlock_ops(slave_dev); 2489 } else { 2490 dev_set_mtu(slave_dev, slave->original_mtu); 2491 } 2492 2493 if (!netif_is_bond_master(slave_dev)) 2494 slave_dev->priv_flags &= ~IFF_BONDING; 2495 2496 bond_xdp_set_features(bond_dev); 2497 kobject_put(&slave->kobj); 2498 2499 return 0; 2500 } 2501 2502 /* A wrapper used because of ndo_del_link */ 2503 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2504 { 2505 return __bond_release_one(bond_dev, slave_dev, false, false); 2506 } 2507 2508 /* First release a slave and then destroy the bond if no more slaves are left. 2509 * Must be under rtnl_lock when this function is called. 2510 */ 2511 static int bond_release_and_destroy(struct net_device *bond_dev, 2512 struct net_device *slave_dev) 2513 { 2514 struct bonding *bond = netdev_priv(bond_dev); 2515 int ret; 2516 2517 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2518 if (ret == 0 && !bond_has_slaves(bond) && 2519 bond_dev->reg_state != NETREG_UNREGISTERING) { 2520 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2521 netdev_info(bond_dev, "Destroying bond\n"); 2522 bond_remove_proc_entry(bond); 2523 unregister_netdevice(bond_dev); 2524 } 2525 return ret; 2526 } 2527 2528 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2529 { 2530 struct bonding *bond = netdev_priv(bond_dev); 2531 2532 bond_fill_ifbond(bond, info); 2533 } 2534 2535 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2536 { 2537 struct bonding *bond = netdev_priv(bond_dev); 2538 struct list_head *iter; 2539 int i = 0, res = -ENODEV; 2540 struct slave *slave; 2541 2542 bond_for_each_slave(bond, slave, iter) { 2543 if (i++ == (int)info->slave_id) { 2544 res = 0; 2545 bond_fill_ifslave(slave, info); 2546 break; 2547 } 2548 } 2549 2550 return res; 2551 } 2552 2553 /*-------------------------------- Monitoring -------------------------------*/ 2554 2555 /* called with rcu_read_lock() */ 2556 static int bond_miimon_inspect(struct bonding *bond) 2557 { 2558 bool ignore_updelay = false; 2559 int link_state, commit = 0; 2560 struct list_head *iter; 2561 struct slave *slave; 2562 2563 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2564 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2565 } else { 2566 struct bond_up_slave *usable_slaves; 2567 2568 usable_slaves = rcu_dereference(bond->usable_slaves); 2569 2570 if (usable_slaves && usable_slaves->count == 0) 2571 ignore_updelay = true; 2572 } 2573 2574 bond_for_each_slave_rcu(bond, slave, iter) { 2575 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2576 2577 link_state = netif_carrier_ok(slave->dev); 2578 2579 switch (slave->link) { 2580 case BOND_LINK_UP: 2581 if (link_state) 2582 continue; 2583 2584 bond_propose_link_state(slave, BOND_LINK_FAIL); 2585 commit++; 2586 slave->delay = bond->params.downdelay; 2587 if (slave->delay && net_ratelimit()) { 2588 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2589 (BOND_MODE(bond) == 2590 BOND_MODE_ACTIVEBACKUP) ? 2591 (bond_is_active_slave(slave) ? 2592 "active " : "backup ") : "", 2593 bond->params.downdelay * bond->params.miimon); 2594 } 2595 fallthrough; 2596 case BOND_LINK_FAIL: 2597 if (link_state) { 2598 /* recovered before downdelay expired */ 2599 bond_propose_link_state(slave, BOND_LINK_UP); 2600 slave->last_link_up = jiffies; 2601 if (net_ratelimit()) 2602 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2603 (bond->params.downdelay - slave->delay) * 2604 bond->params.miimon); 2605 commit++; 2606 continue; 2607 } 2608 2609 if (slave->delay <= 0) { 2610 bond_propose_link_state(slave, BOND_LINK_DOWN); 2611 commit++; 2612 continue; 2613 } 2614 2615 slave->delay--; 2616 break; 2617 2618 case BOND_LINK_DOWN: 2619 if (!link_state) 2620 continue; 2621 2622 bond_propose_link_state(slave, BOND_LINK_BACK); 2623 commit++; 2624 slave->delay = bond->params.updelay; 2625 2626 if (slave->delay && net_ratelimit()) { 2627 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2628 ignore_updelay ? 0 : 2629 bond->params.updelay * 2630 bond->params.miimon); 2631 } 2632 fallthrough; 2633 case BOND_LINK_BACK: 2634 if (!link_state) { 2635 bond_propose_link_state(slave, BOND_LINK_DOWN); 2636 if (net_ratelimit()) 2637 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2638 (bond->params.updelay - slave->delay) * 2639 bond->params.miimon); 2640 commit++; 2641 continue; 2642 } 2643 2644 if (ignore_updelay) 2645 slave->delay = 0; 2646 2647 if (slave->delay <= 0) { 2648 bond_propose_link_state(slave, BOND_LINK_UP); 2649 commit++; 2650 ignore_updelay = false; 2651 continue; 2652 } 2653 2654 slave->delay--; 2655 break; 2656 } 2657 } 2658 2659 return commit; 2660 } 2661 2662 static void bond_miimon_link_change(struct bonding *bond, 2663 struct slave *slave, 2664 char link) 2665 { 2666 switch (BOND_MODE(bond)) { 2667 case BOND_MODE_8023AD: 2668 bond_3ad_handle_link_change(slave, link); 2669 break; 2670 case BOND_MODE_TLB: 2671 case BOND_MODE_ALB: 2672 bond_alb_handle_link_change(bond, slave, link); 2673 break; 2674 case BOND_MODE_XOR: 2675 bond_update_slave_arr(bond, NULL); 2676 break; 2677 } 2678 } 2679 2680 static void bond_miimon_commit(struct bonding *bond) 2681 { 2682 struct slave *slave, *primary, *active; 2683 bool do_failover = false; 2684 struct list_head *iter; 2685 2686 ASSERT_RTNL(); 2687 2688 bond_for_each_slave(bond, slave, iter) { 2689 switch (slave->link_new_state) { 2690 case BOND_LINK_NOCHANGE: 2691 /* For 802.3ad mode, check current slave speed and 2692 * duplex again in case its port was disabled after 2693 * invalid speed/duplex reporting but recovered before 2694 * link monitoring could make a decision on the actual 2695 * link status 2696 */ 2697 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2698 slave->link == BOND_LINK_UP) 2699 bond_3ad_adapter_speed_duplex_changed(slave); 2700 continue; 2701 2702 case BOND_LINK_UP: 2703 if (bond_update_speed_duplex(slave) && 2704 bond_needs_speed_duplex(bond)) { 2705 slave->link = BOND_LINK_DOWN; 2706 if (net_ratelimit()) 2707 slave_warn(bond->dev, slave->dev, 2708 "failed to get link speed/duplex\n"); 2709 continue; 2710 } 2711 bond_set_slave_link_state(slave, BOND_LINK_UP, 2712 BOND_SLAVE_NOTIFY_NOW); 2713 slave->last_link_up = jiffies; 2714 2715 primary = rtnl_dereference(bond->primary_slave); 2716 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2717 /* prevent it from being the active one */ 2718 bond_set_backup_slave(slave); 2719 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2720 /* make it immediately active */ 2721 bond_set_active_slave(slave); 2722 } 2723 2724 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2725 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2726 slave->duplex ? "full" : "half"); 2727 2728 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2729 2730 active = rtnl_dereference(bond->curr_active_slave); 2731 if (!active || slave == primary || slave->prio > active->prio) 2732 do_failover = true; 2733 2734 continue; 2735 2736 case BOND_LINK_DOWN: 2737 if (slave->link_failure_count < UINT_MAX) 2738 slave->link_failure_count++; 2739 2740 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2741 BOND_SLAVE_NOTIFY_NOW); 2742 2743 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2744 BOND_MODE(bond) == BOND_MODE_8023AD) 2745 bond_set_slave_inactive_flags(slave, 2746 BOND_SLAVE_NOTIFY_NOW); 2747 2748 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2749 2750 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2751 2752 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2753 do_failover = true; 2754 2755 continue; 2756 2757 default: 2758 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2759 slave->link_new_state); 2760 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2761 2762 continue; 2763 } 2764 } 2765 2766 if (do_failover) { 2767 block_netpoll_tx(); 2768 bond_select_active_slave(bond); 2769 unblock_netpoll_tx(); 2770 } 2771 2772 bond_set_carrier(bond); 2773 } 2774 2775 /* bond_mii_monitor 2776 * 2777 * Really a wrapper that splits the mii monitor into two phases: an 2778 * inspection, then (if inspection indicates something needs to be done) 2779 * an acquisition of appropriate locks followed by a commit phase to 2780 * implement whatever link state changes are indicated. 2781 */ 2782 static void bond_mii_monitor(struct work_struct *work) 2783 { 2784 struct bonding *bond = container_of(work, struct bonding, 2785 mii_work.work); 2786 bool should_notify_peers; 2787 bool commit; 2788 unsigned long delay; 2789 struct slave *slave; 2790 struct list_head *iter; 2791 2792 delay = msecs_to_jiffies(bond->params.miimon); 2793 2794 if (!bond_has_slaves(bond)) 2795 goto re_arm; 2796 2797 rcu_read_lock(); 2798 2799 should_notify_peers = bond_should_notify_peers(bond); 2800 commit = !!bond_miimon_inspect(bond); 2801 2802 rcu_read_unlock(); 2803 2804 if (commit || bond->send_peer_notif) { 2805 /* Race avoidance with bond_close cancel of workqueue */ 2806 if (!rtnl_trylock()) { 2807 delay = 1; 2808 goto re_arm; 2809 } 2810 2811 if (commit) { 2812 bond_for_each_slave(bond, slave, iter) { 2813 bond_commit_link_state(slave, 2814 BOND_SLAVE_NOTIFY_LATER); 2815 } 2816 bond_miimon_commit(bond); 2817 } 2818 2819 if (bond->send_peer_notif) { 2820 bond->send_peer_notif--; 2821 if (should_notify_peers) 2822 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 2823 bond->dev); 2824 } 2825 2826 rtnl_unlock(); /* might sleep, hold no other locks */ 2827 } 2828 2829 re_arm: 2830 if (bond->params.miimon) 2831 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2832 } 2833 2834 static int bond_upper_dev_walk(struct net_device *upper, 2835 struct netdev_nested_priv *priv) 2836 { 2837 __be32 ip = *(__be32 *)priv->data; 2838 2839 return ip == bond_confirm_addr(upper, 0, ip); 2840 } 2841 2842 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2843 { 2844 struct netdev_nested_priv priv = { 2845 .data = (void *)&ip, 2846 }; 2847 bool ret = false; 2848 2849 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2850 return true; 2851 2852 rcu_read_lock(); 2853 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2854 ret = true; 2855 rcu_read_unlock(); 2856 2857 return ret; 2858 } 2859 2860 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff) 2861 2862 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2863 struct sk_buff *skb) 2864 { 2865 struct net_device *bond_dev = slave->bond->dev; 2866 struct net_device *slave_dev = slave->dev; 2867 struct bond_vlan_tag *outer_tag = tags; 2868 2869 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE) 2870 return true; 2871 2872 tags++; 2873 2874 /* Go through all the tags backwards and add them to the packet */ 2875 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) { 2876 if (!tags->vlan_id) { 2877 tags++; 2878 continue; 2879 } 2880 2881 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2882 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2883 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2884 tags->vlan_id); 2885 if (!skb) { 2886 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2887 return false; 2888 } 2889 2890 tags++; 2891 } 2892 /* Set the outer tag */ 2893 if (outer_tag->vlan_id) { 2894 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2895 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2896 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2897 outer_tag->vlan_id); 2898 } 2899 2900 return true; 2901 } 2902 2903 /* We go to the (large) trouble of VLAN tagging ARP frames because 2904 * switches in VLAN mode (especially if ports are configured as 2905 * "native" to a VLAN) might not pass non-tagged frames. 2906 */ 2907 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2908 __be32 src_ip, struct bond_vlan_tag *tags) 2909 { 2910 struct net_device *bond_dev = slave->bond->dev; 2911 struct net_device *slave_dev = slave->dev; 2912 struct sk_buff *skb; 2913 2914 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2915 arp_op, &dest_ip, &src_ip); 2916 2917 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2918 NULL, slave_dev->dev_addr, NULL); 2919 2920 if (!skb) { 2921 net_err_ratelimited("ARP packet allocation failed\n"); 2922 return; 2923 } 2924 2925 if (bond_handle_vlan(slave, tags, skb)) { 2926 slave_update_last_tx(slave); 2927 arp_xmit(skb); 2928 } 2929 2930 return; 2931 } 2932 2933 /* Validate the device path between the @start_dev and the @end_dev. 2934 * The path is valid if the @end_dev is reachable through device 2935 * stacking. 2936 * When the path is validated, collect any vlan information in the 2937 * path. 2938 */ 2939 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2940 struct net_device *end_dev, 2941 int level) 2942 { 2943 struct bond_vlan_tag *tags; 2944 struct net_device *upper; 2945 struct list_head *iter; 2946 2947 if (start_dev == end_dev) { 2948 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2949 if (!tags) 2950 return ERR_PTR(-ENOMEM); 2951 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE; 2952 return tags; 2953 } 2954 2955 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2956 tags = bond_verify_device_path(upper, end_dev, level + 1); 2957 if (IS_ERR_OR_NULL(tags)) { 2958 if (IS_ERR(tags)) 2959 return tags; 2960 continue; 2961 } 2962 if (is_vlan_dev(upper)) { 2963 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2964 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2965 } 2966 2967 return tags; 2968 } 2969 2970 return NULL; 2971 } 2972 2973 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2974 { 2975 struct rtable *rt; 2976 struct bond_vlan_tag *tags; 2977 __be32 *targets = bond->params.arp_targets, addr; 2978 int i; 2979 2980 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2981 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2982 __func__, &targets[i]); 2983 tags = NULL; 2984 2985 /* Find out through which dev should the packet go */ 2986 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0, 2987 RT_SCOPE_LINK); 2988 if (IS_ERR(rt)) { 2989 /* there's no route to target - try to send arp 2990 * probe to generate any traffic (arp_validate=0) 2991 */ 2992 if (bond->params.arp_validate) 2993 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2994 bond->dev->name, 2995 &targets[i]); 2996 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2997 0, tags); 2998 continue; 2999 } 3000 3001 /* bond device itself */ 3002 if (rt->dst.dev == bond->dev) 3003 goto found; 3004 3005 rcu_read_lock(); 3006 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 3007 rcu_read_unlock(); 3008 3009 if (!IS_ERR_OR_NULL(tags)) 3010 goto found; 3011 3012 /* Not our device - skip */ 3013 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 3014 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 3015 3016 ip_rt_put(rt); 3017 continue; 3018 3019 found: 3020 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 3021 ip_rt_put(rt); 3022 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 3023 kfree(tags); 3024 } 3025 } 3026 3027 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 3028 { 3029 int i; 3030 3031 if (!sip || !bond_has_this_ip(bond, tip)) { 3032 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 3033 __func__, &sip, &tip); 3034 return; 3035 } 3036 3037 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3038 if (i == -1) { 3039 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3040 __func__, &sip); 3041 return; 3042 } 3043 slave->last_rx = jiffies; 3044 slave->target_last_arp_rx[i] = jiffies; 3045 } 3046 3047 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3048 struct slave *slave) 3049 { 3050 struct arphdr *arp = (struct arphdr *)skb->data; 3051 struct slave *curr_active_slave, *curr_arp_slave; 3052 unsigned char *arp_ptr; 3053 __be32 sip, tip; 3054 unsigned int alen; 3055 3056 alen = arp_hdr_len(bond->dev); 3057 3058 if (alen > skb_headlen(skb)) { 3059 arp = kmalloc(alen, GFP_ATOMIC); 3060 if (!arp) 3061 goto out_unlock; 3062 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3063 goto out_unlock; 3064 } 3065 3066 if (arp->ar_hln != bond->dev->addr_len || 3067 skb->pkt_type == PACKET_OTHERHOST || 3068 skb->pkt_type == PACKET_LOOPBACK || 3069 arp->ar_hrd != htons(ARPHRD_ETHER) || 3070 arp->ar_pro != htons(ETH_P_IP) || 3071 arp->ar_pln != 4) 3072 goto out_unlock; 3073 3074 arp_ptr = (unsigned char *)(arp + 1); 3075 arp_ptr += bond->dev->addr_len; 3076 memcpy(&sip, arp_ptr, 4); 3077 arp_ptr += 4 + bond->dev->addr_len; 3078 memcpy(&tip, arp_ptr, 4); 3079 3080 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3081 __func__, slave->dev->name, bond_slave_state(slave), 3082 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3083 &sip, &tip); 3084 3085 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3086 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3087 3088 /* We 'trust' the received ARP enough to validate it if: 3089 * 3090 * (a) the slave receiving the ARP is active (which includes the 3091 * current ARP slave, if any), or 3092 * 3093 * (b) the receiving slave isn't active, but there is a currently 3094 * active slave and it received valid arp reply(s) after it became 3095 * the currently active slave, or 3096 * 3097 * (c) there is an ARP slave that sent an ARP during the prior ARP 3098 * interval, and we receive an ARP reply on any slave. We accept 3099 * these because switch FDB update delays may deliver the ARP 3100 * reply to a slave other than the sender of the ARP request. 3101 * 3102 * Note: for (b), backup slaves are receiving the broadcast ARP 3103 * request, not a reply. This request passes from the sending 3104 * slave through the L2 switch(es) to the receiving slave. Since 3105 * this is checking the request, sip/tip are swapped for 3106 * validation. 3107 * 3108 * This is done to avoid endless looping when we can't reach the 3109 * arp_ip_target and fool ourselves with our own arp requests. 3110 */ 3111 if (bond_is_active_slave(slave)) 3112 bond_validate_arp(bond, slave, sip, tip); 3113 else if (curr_active_slave && 3114 time_after(slave_last_rx(bond, curr_active_slave), 3115 curr_active_slave->last_link_up)) 3116 bond_validate_arp(bond, slave, tip, sip); 3117 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3118 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3119 bond_validate_arp(bond, slave, sip, tip); 3120 3121 out_unlock: 3122 if (arp != (struct arphdr *)skb->data) 3123 kfree(arp); 3124 return RX_HANDLER_ANOTHER; 3125 } 3126 3127 #if IS_ENABLED(CONFIG_IPV6) 3128 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3129 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3130 { 3131 struct net_device *bond_dev = slave->bond->dev; 3132 struct net_device *slave_dev = slave->dev; 3133 struct in6_addr mcaddr; 3134 struct sk_buff *skb; 3135 3136 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3137 daddr, saddr); 3138 3139 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3140 if (!skb) { 3141 net_err_ratelimited("NS packet allocation failed\n"); 3142 return; 3143 } 3144 3145 addrconf_addr_solict_mult(daddr, &mcaddr); 3146 if (bond_handle_vlan(slave, tags, skb)) { 3147 slave_update_last_tx(slave); 3148 ndisc_send_skb(skb, &mcaddr, saddr); 3149 } 3150 } 3151 3152 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3153 { 3154 struct in6_addr *targets = bond->params.ns_targets; 3155 struct bond_vlan_tag *tags; 3156 struct dst_entry *dst; 3157 struct in6_addr saddr; 3158 struct flowi6 fl6; 3159 int i; 3160 3161 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3162 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3163 __func__, &targets[i]); 3164 tags = NULL; 3165 3166 /* Find out through which dev should the packet go */ 3167 memset(&fl6, 0, sizeof(struct flowi6)); 3168 fl6.daddr = targets[i]; 3169 3170 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3171 if (dst->error) { 3172 dst_release(dst); 3173 /* there's no route to target - try to send arp 3174 * probe to generate any traffic (arp_validate=0) 3175 */ 3176 if (bond->params.arp_validate) 3177 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3178 bond->dev->name, 3179 &targets[i]); 3180 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3181 continue; 3182 } 3183 3184 /* bond device itself */ 3185 if (dst->dev == bond->dev) 3186 goto found; 3187 3188 rcu_read_lock(); 3189 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3190 rcu_read_unlock(); 3191 3192 if (!IS_ERR_OR_NULL(tags)) 3193 goto found; 3194 3195 /* Not our device - skip */ 3196 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3197 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3198 3199 dst_release(dst); 3200 continue; 3201 3202 found: 3203 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3204 bond_ns_send(slave, &targets[i], &saddr, tags); 3205 else 3206 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3207 3208 dst_release(dst); 3209 kfree(tags); 3210 } 3211 } 3212 3213 static int bond_confirm_addr6(struct net_device *dev, 3214 struct netdev_nested_priv *priv) 3215 { 3216 struct in6_addr *addr = (struct in6_addr *)priv->data; 3217 3218 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3219 } 3220 3221 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3222 { 3223 struct netdev_nested_priv priv = { 3224 .data = addr, 3225 }; 3226 int ret = false; 3227 3228 if (bond_confirm_addr6(bond->dev, &priv)) 3229 return true; 3230 3231 rcu_read_lock(); 3232 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3233 ret = true; 3234 rcu_read_unlock(); 3235 3236 return ret; 3237 } 3238 3239 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3240 struct in6_addr *saddr, struct in6_addr *daddr) 3241 { 3242 int i; 3243 3244 /* Ignore NAs that: 3245 * 1. Source address is unspecified address. 3246 * 2. Dest address is neither all-nodes multicast address nor 3247 * exist on bond interface. 3248 */ 3249 if (ipv6_addr_any(saddr) || 3250 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3251 !bond_has_this_ip6(bond, daddr))) { 3252 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3253 __func__, saddr, daddr); 3254 return; 3255 } 3256 3257 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3258 if (i == -1) { 3259 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3260 __func__, saddr); 3261 return; 3262 } 3263 slave->last_rx = jiffies; 3264 slave->target_last_arp_rx[i] = jiffies; 3265 } 3266 3267 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3268 struct slave *slave) 3269 { 3270 struct slave *curr_active_slave, *curr_arp_slave; 3271 struct in6_addr *saddr, *daddr; 3272 struct { 3273 struct ipv6hdr ip6; 3274 struct icmp6hdr icmp6; 3275 } *combined, _combined; 3276 3277 if (skb->pkt_type == PACKET_OTHERHOST || 3278 skb->pkt_type == PACKET_LOOPBACK) 3279 goto out; 3280 3281 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3282 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3283 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION && 3284 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)) 3285 goto out; 3286 3287 saddr = &combined->ip6.saddr; 3288 daddr = &combined->ip6.daddr; 3289 3290 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3291 __func__, slave->dev->name, bond_slave_state(slave), 3292 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3293 saddr, daddr); 3294 3295 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3296 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3297 3298 /* We 'trust' the received ARP enough to validate it if: 3299 * see bond_arp_rcv(). 3300 */ 3301 if (bond_is_active_slave(slave)) 3302 bond_validate_na(bond, slave, saddr, daddr); 3303 else if (curr_active_slave && 3304 time_after(slave_last_rx(bond, curr_active_slave), 3305 curr_active_slave->last_link_up)) 3306 bond_validate_na(bond, slave, daddr, saddr); 3307 else if (curr_arp_slave && 3308 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3309 bond_validate_na(bond, slave, saddr, daddr); 3310 3311 out: 3312 return RX_HANDLER_ANOTHER; 3313 } 3314 #endif 3315 3316 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3317 struct slave *slave) 3318 { 3319 #if IS_ENABLED(CONFIG_IPV6) 3320 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3321 #endif 3322 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3323 3324 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3325 __func__, skb->dev->name); 3326 3327 /* Use arp validate logic for both ARP and NS */ 3328 if (!slave_do_arp_validate(bond, slave)) { 3329 if ((slave_do_arp_validate_only(bond) && is_arp) || 3330 #if IS_ENABLED(CONFIG_IPV6) 3331 (slave_do_arp_validate_only(bond) && is_ipv6) || 3332 #endif 3333 !slave_do_arp_validate_only(bond)) 3334 slave->last_rx = jiffies; 3335 return RX_HANDLER_ANOTHER; 3336 } else if (is_arp) { 3337 return bond_arp_rcv(skb, bond, slave); 3338 #if IS_ENABLED(CONFIG_IPV6) 3339 } else if (is_ipv6) { 3340 return bond_na_rcv(skb, bond, slave); 3341 #endif 3342 } else { 3343 return RX_HANDLER_ANOTHER; 3344 } 3345 } 3346 3347 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3348 { 3349 bond_arp_send_all(bond, slave); 3350 #if IS_ENABLED(CONFIG_IPV6) 3351 bond_ns_send_all(bond, slave); 3352 #endif 3353 } 3354 3355 /* function to verify if we're in the arp_interval timeslice, returns true if 3356 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3357 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3358 */ 3359 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3360 int mod) 3361 { 3362 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3363 3364 return time_in_range(jiffies, 3365 last_act - delta_in_ticks, 3366 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3367 } 3368 3369 /* This function is called regularly to monitor each slave's link 3370 * ensuring that traffic is being sent and received when arp monitoring 3371 * is used in load-balancing mode. if the adapter has been dormant, then an 3372 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3373 * arp monitoring in active backup mode. 3374 */ 3375 static void bond_loadbalance_arp_mon(struct bonding *bond) 3376 { 3377 struct slave *slave, *oldcurrent; 3378 struct list_head *iter; 3379 int do_failover = 0, slave_state_changed = 0; 3380 3381 if (!bond_has_slaves(bond)) 3382 goto re_arm; 3383 3384 rcu_read_lock(); 3385 3386 oldcurrent = rcu_dereference(bond->curr_active_slave); 3387 /* see if any of the previous devices are up now (i.e. they have 3388 * xmt and rcv traffic). the curr_active_slave does not come into 3389 * the picture unless it is null. also, slave->last_link_up is not 3390 * needed here because we send an arp on each slave and give a slave 3391 * as long as it needs to get the tx/rx within the delta. 3392 * TODO: what about up/down delay in arp mode? it wasn't here before 3393 * so it can wait 3394 */ 3395 bond_for_each_slave_rcu(bond, slave, iter) { 3396 unsigned long last_tx = slave_last_tx(slave); 3397 3398 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3399 3400 if (slave->link != BOND_LINK_UP) { 3401 if (bond_time_in_interval(bond, last_tx, 1) && 3402 bond_time_in_interval(bond, slave->last_rx, 1)) { 3403 3404 bond_propose_link_state(slave, BOND_LINK_UP); 3405 slave_state_changed = 1; 3406 3407 /* primary_slave has no meaning in round-robin 3408 * mode. the window of a slave being up and 3409 * curr_active_slave being null after enslaving 3410 * is closed. 3411 */ 3412 if (!oldcurrent) { 3413 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3414 do_failover = 1; 3415 } else { 3416 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3417 } 3418 } 3419 } else { 3420 /* slave->link == BOND_LINK_UP */ 3421 3422 /* not all switches will respond to an arp request 3423 * when the source ip is 0, so don't take the link down 3424 * if we don't know our ip yet 3425 */ 3426 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3427 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3428 3429 bond_propose_link_state(slave, BOND_LINK_DOWN); 3430 slave_state_changed = 1; 3431 3432 if (slave->link_failure_count < UINT_MAX) 3433 slave->link_failure_count++; 3434 3435 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3436 3437 if (slave == oldcurrent) 3438 do_failover = 1; 3439 } 3440 } 3441 3442 /* note: if switch is in round-robin mode, all links 3443 * must tx arp to ensure all links rx an arp - otherwise 3444 * links may oscillate or not come up at all; if switch is 3445 * in something like xor mode, there is nothing we can 3446 * do - all replies will be rx'ed on same link causing slaves 3447 * to be unstable during low/no traffic periods 3448 */ 3449 if (bond_slave_is_up(slave)) 3450 bond_send_validate(bond, slave); 3451 } 3452 3453 rcu_read_unlock(); 3454 3455 if (do_failover || slave_state_changed) { 3456 if (!rtnl_trylock()) 3457 goto re_arm; 3458 3459 bond_for_each_slave(bond, slave, iter) { 3460 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3461 slave->link = slave->link_new_state; 3462 } 3463 3464 if (slave_state_changed) { 3465 bond_slave_state_change(bond); 3466 if (BOND_MODE(bond) == BOND_MODE_XOR) 3467 bond_update_slave_arr(bond, NULL); 3468 } 3469 if (do_failover) { 3470 block_netpoll_tx(); 3471 bond_select_active_slave(bond); 3472 unblock_netpoll_tx(); 3473 } 3474 rtnl_unlock(); 3475 } 3476 3477 re_arm: 3478 if (bond->params.arp_interval) 3479 queue_delayed_work(bond->wq, &bond->arp_work, 3480 msecs_to_jiffies(bond->params.arp_interval)); 3481 } 3482 3483 /* Called to inspect slaves for active-backup mode ARP monitor link state 3484 * changes. Sets proposed link state in slaves to specify what action 3485 * should take place for the slave. Returns 0 if no changes are found, >0 3486 * if changes to link states must be committed. 3487 * 3488 * Called with rcu_read_lock held. 3489 */ 3490 static int bond_ab_arp_inspect(struct bonding *bond) 3491 { 3492 unsigned long last_tx, last_rx; 3493 struct list_head *iter; 3494 struct slave *slave; 3495 int commit = 0; 3496 3497 bond_for_each_slave_rcu(bond, slave, iter) { 3498 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3499 last_rx = slave_last_rx(bond, slave); 3500 3501 if (slave->link != BOND_LINK_UP) { 3502 if (bond_time_in_interval(bond, last_rx, 1)) { 3503 bond_propose_link_state(slave, BOND_LINK_UP); 3504 commit++; 3505 } else if (slave->link == BOND_LINK_BACK) { 3506 bond_propose_link_state(slave, BOND_LINK_FAIL); 3507 commit++; 3508 } 3509 continue; 3510 } 3511 3512 /* Give slaves 2*delta after being enslaved or made 3513 * active. This avoids bouncing, as the last receive 3514 * times need a full ARP monitor cycle to be updated. 3515 */ 3516 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3517 continue; 3518 3519 /* Backup slave is down if: 3520 * - No current_arp_slave AND 3521 * - more than (missed_max+1)*delta since last receive AND 3522 * - the bond has an IP address 3523 * 3524 * Note: a non-null current_arp_slave indicates 3525 * the curr_active_slave went down and we are 3526 * searching for a new one; under this condition 3527 * we only take the curr_active_slave down - this 3528 * gives each slave a chance to tx/rx traffic 3529 * before being taken out 3530 */ 3531 if (!bond_is_active_slave(slave) && 3532 !rcu_access_pointer(bond->current_arp_slave) && 3533 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3534 bond_propose_link_state(slave, BOND_LINK_DOWN); 3535 commit++; 3536 } 3537 3538 /* Active slave is down if: 3539 * - more than missed_max*delta since transmitting OR 3540 * - (more than missed_max*delta since receive AND 3541 * the bond has an IP address) 3542 */ 3543 last_tx = slave_last_tx(slave); 3544 if (bond_is_active_slave(slave) && 3545 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3546 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3547 bond_propose_link_state(slave, BOND_LINK_DOWN); 3548 commit++; 3549 } 3550 } 3551 3552 return commit; 3553 } 3554 3555 /* Called to commit link state changes noted by inspection step of 3556 * active-backup mode ARP monitor. 3557 * 3558 * Called with RTNL hold. 3559 */ 3560 static void bond_ab_arp_commit(struct bonding *bond) 3561 { 3562 bool do_failover = false; 3563 struct list_head *iter; 3564 unsigned long last_tx; 3565 struct slave *slave; 3566 3567 bond_for_each_slave(bond, slave, iter) { 3568 switch (slave->link_new_state) { 3569 case BOND_LINK_NOCHANGE: 3570 continue; 3571 3572 case BOND_LINK_UP: 3573 last_tx = slave_last_tx(slave); 3574 if (rtnl_dereference(bond->curr_active_slave) != slave || 3575 (!rtnl_dereference(bond->curr_active_slave) && 3576 bond_time_in_interval(bond, last_tx, 1))) { 3577 struct slave *current_arp_slave; 3578 3579 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3580 bond_set_slave_link_state(slave, BOND_LINK_UP, 3581 BOND_SLAVE_NOTIFY_NOW); 3582 if (current_arp_slave) { 3583 bond_set_slave_inactive_flags( 3584 current_arp_slave, 3585 BOND_SLAVE_NOTIFY_NOW); 3586 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3587 } 3588 3589 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3590 3591 if (!rtnl_dereference(bond->curr_active_slave) || 3592 slave == rtnl_dereference(bond->primary_slave) || 3593 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio) 3594 do_failover = true; 3595 3596 } 3597 3598 continue; 3599 3600 case BOND_LINK_DOWN: 3601 if (slave->link_failure_count < UINT_MAX) 3602 slave->link_failure_count++; 3603 3604 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3605 BOND_SLAVE_NOTIFY_NOW); 3606 bond_set_slave_inactive_flags(slave, 3607 BOND_SLAVE_NOTIFY_NOW); 3608 3609 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3610 3611 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3612 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3613 do_failover = true; 3614 } 3615 3616 continue; 3617 3618 case BOND_LINK_FAIL: 3619 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3620 BOND_SLAVE_NOTIFY_NOW); 3621 bond_set_slave_inactive_flags(slave, 3622 BOND_SLAVE_NOTIFY_NOW); 3623 3624 /* A slave has just been enslaved and has become 3625 * the current active slave. 3626 */ 3627 if (rtnl_dereference(bond->curr_active_slave)) 3628 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3629 continue; 3630 3631 default: 3632 slave_err(bond->dev, slave->dev, 3633 "impossible: link_new_state %d on slave\n", 3634 slave->link_new_state); 3635 continue; 3636 } 3637 } 3638 3639 if (do_failover) { 3640 block_netpoll_tx(); 3641 bond_select_active_slave(bond); 3642 unblock_netpoll_tx(); 3643 } 3644 3645 bond_set_carrier(bond); 3646 } 3647 3648 /* Send ARP probes for active-backup mode ARP monitor. 3649 * 3650 * Called with rcu_read_lock held. 3651 */ 3652 static bool bond_ab_arp_probe(struct bonding *bond) 3653 { 3654 struct slave *slave, *before = NULL, *new_slave = NULL, 3655 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3656 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3657 struct list_head *iter; 3658 bool found = false; 3659 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3660 3661 if (curr_arp_slave && curr_active_slave) 3662 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3663 curr_arp_slave->dev->name, 3664 curr_active_slave->dev->name); 3665 3666 if (curr_active_slave) { 3667 bond_send_validate(bond, curr_active_slave); 3668 return should_notify_rtnl; 3669 } 3670 3671 /* if we don't have a curr_active_slave, search for the next available 3672 * backup slave from the current_arp_slave and make it the candidate 3673 * for becoming the curr_active_slave 3674 */ 3675 3676 if (!curr_arp_slave) { 3677 curr_arp_slave = bond_first_slave_rcu(bond); 3678 if (!curr_arp_slave) 3679 return should_notify_rtnl; 3680 } 3681 3682 bond_for_each_slave_rcu(bond, slave, iter) { 3683 if (!found && !before && bond_slave_is_up(slave)) 3684 before = slave; 3685 3686 if (found && !new_slave && bond_slave_is_up(slave)) 3687 new_slave = slave; 3688 /* if the link state is up at this point, we 3689 * mark it down - this can happen if we have 3690 * simultaneous link failures and 3691 * reselect_active_interface doesn't make this 3692 * one the current slave so it is still marked 3693 * up when it is actually down 3694 */ 3695 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3696 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3697 BOND_SLAVE_NOTIFY_LATER); 3698 if (slave->link_failure_count < UINT_MAX) 3699 slave->link_failure_count++; 3700 3701 bond_set_slave_inactive_flags(slave, 3702 BOND_SLAVE_NOTIFY_LATER); 3703 3704 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3705 } 3706 if (slave == curr_arp_slave) 3707 found = true; 3708 } 3709 3710 if (!new_slave && before) 3711 new_slave = before; 3712 3713 if (!new_slave) 3714 goto check_state; 3715 3716 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3717 BOND_SLAVE_NOTIFY_LATER); 3718 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3719 bond_send_validate(bond, new_slave); 3720 new_slave->last_link_up = jiffies; 3721 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3722 3723 check_state: 3724 bond_for_each_slave_rcu(bond, slave, iter) { 3725 if (slave->should_notify || slave->should_notify_link) { 3726 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3727 break; 3728 } 3729 } 3730 return should_notify_rtnl; 3731 } 3732 3733 static void bond_activebackup_arp_mon(struct bonding *bond) 3734 { 3735 bool should_notify_peers = false; 3736 bool should_notify_rtnl = false; 3737 int delta_in_ticks; 3738 3739 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3740 3741 if (!bond_has_slaves(bond)) 3742 goto re_arm; 3743 3744 rcu_read_lock(); 3745 3746 should_notify_peers = bond_should_notify_peers(bond); 3747 3748 if (bond_ab_arp_inspect(bond)) { 3749 rcu_read_unlock(); 3750 3751 /* Race avoidance with bond_close flush of workqueue */ 3752 if (!rtnl_trylock()) { 3753 delta_in_ticks = 1; 3754 should_notify_peers = false; 3755 goto re_arm; 3756 } 3757 3758 bond_ab_arp_commit(bond); 3759 3760 rtnl_unlock(); 3761 rcu_read_lock(); 3762 } 3763 3764 should_notify_rtnl = bond_ab_arp_probe(bond); 3765 rcu_read_unlock(); 3766 3767 re_arm: 3768 if (bond->params.arp_interval) 3769 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3770 3771 if (should_notify_peers || should_notify_rtnl) { 3772 if (!rtnl_trylock()) 3773 return; 3774 3775 if (should_notify_peers) { 3776 bond->send_peer_notif--; 3777 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3778 bond->dev); 3779 } 3780 if (should_notify_rtnl) { 3781 bond_slave_state_notify(bond); 3782 bond_slave_link_notify(bond); 3783 } 3784 3785 rtnl_unlock(); 3786 } 3787 } 3788 3789 static void bond_arp_monitor(struct work_struct *work) 3790 { 3791 struct bonding *bond = container_of(work, struct bonding, 3792 arp_work.work); 3793 3794 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3795 bond_activebackup_arp_mon(bond); 3796 else 3797 bond_loadbalance_arp_mon(bond); 3798 } 3799 3800 /*-------------------------- netdev event handling --------------------------*/ 3801 3802 /* Change device name */ 3803 static int bond_event_changename(struct bonding *bond) 3804 { 3805 bond_remove_proc_entry(bond); 3806 bond_create_proc_entry(bond); 3807 3808 bond_debug_reregister(bond); 3809 3810 return NOTIFY_DONE; 3811 } 3812 3813 static int bond_master_netdev_event(unsigned long event, 3814 struct net_device *bond_dev) 3815 { 3816 struct bonding *event_bond = netdev_priv(bond_dev); 3817 3818 netdev_dbg(bond_dev, "%s called\n", __func__); 3819 3820 switch (event) { 3821 case NETDEV_CHANGENAME: 3822 return bond_event_changename(event_bond); 3823 case NETDEV_UNREGISTER: 3824 bond_remove_proc_entry(event_bond); 3825 #ifdef CONFIG_XFRM_OFFLOAD 3826 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3827 #endif /* CONFIG_XFRM_OFFLOAD */ 3828 break; 3829 case NETDEV_REGISTER: 3830 bond_create_proc_entry(event_bond); 3831 break; 3832 default: 3833 break; 3834 } 3835 3836 return NOTIFY_DONE; 3837 } 3838 3839 static int bond_slave_netdev_event(unsigned long event, 3840 struct net_device *slave_dev) 3841 { 3842 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3843 struct bonding *bond; 3844 struct net_device *bond_dev; 3845 3846 /* A netdev event can be generated while enslaving a device 3847 * before netdev_rx_handler_register is called in which case 3848 * slave will be NULL 3849 */ 3850 if (!slave) { 3851 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3852 return NOTIFY_DONE; 3853 } 3854 3855 bond_dev = slave->bond->dev; 3856 bond = slave->bond; 3857 primary = rtnl_dereference(bond->primary_slave); 3858 3859 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3860 3861 switch (event) { 3862 case NETDEV_UNREGISTER: 3863 if (bond_dev->type != ARPHRD_ETHER) 3864 bond_release_and_destroy(bond_dev, slave_dev); 3865 else 3866 __bond_release_one(bond_dev, slave_dev, false, true); 3867 break; 3868 case NETDEV_UP: 3869 case NETDEV_CHANGE: 3870 /* For 802.3ad mode only: 3871 * Getting invalid Speed/Duplex values here will put slave 3872 * in weird state. Mark it as link-fail if the link was 3873 * previously up or link-down if it hasn't yet come up, and 3874 * let link-monitoring (miimon) set it right when correct 3875 * speeds/duplex are available. 3876 */ 3877 if (bond_update_speed_duplex(slave) && 3878 BOND_MODE(bond) == BOND_MODE_8023AD) { 3879 if (slave->last_link_up) 3880 slave->link = BOND_LINK_FAIL; 3881 else 3882 slave->link = BOND_LINK_DOWN; 3883 } 3884 3885 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3886 bond_3ad_adapter_speed_duplex_changed(slave); 3887 fallthrough; 3888 case NETDEV_DOWN: 3889 /* Refresh slave-array if applicable! 3890 * If the setup does not use miimon or arpmon (mode-specific!), 3891 * then these events will not cause the slave-array to be 3892 * refreshed. This will cause xmit to use a slave that is not 3893 * usable. Avoid such situation by refeshing the array at these 3894 * events. If these (miimon/arpmon) parameters are configured 3895 * then array gets refreshed twice and that should be fine! 3896 */ 3897 if (bond_mode_can_use_xmit_hash(bond)) 3898 bond_update_slave_arr(bond, NULL); 3899 break; 3900 case NETDEV_CHANGEMTU: 3901 /* TODO: Should slaves be allowed to 3902 * independently alter their MTU? For 3903 * an active-backup bond, slaves need 3904 * not be the same type of device, so 3905 * MTUs may vary. For other modes, 3906 * slaves arguably should have the 3907 * same MTUs. To do this, we'd need to 3908 * take over the slave's change_mtu 3909 * function for the duration of their 3910 * servitude. 3911 */ 3912 break; 3913 case NETDEV_CHANGENAME: 3914 /* we don't care if we don't have primary set */ 3915 if (!bond_uses_primary(bond) || 3916 !bond->params.primary[0]) 3917 break; 3918 3919 if (slave == primary) { 3920 /* slave's name changed - he's no longer primary */ 3921 RCU_INIT_POINTER(bond->primary_slave, NULL); 3922 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3923 /* we have a new primary slave */ 3924 rcu_assign_pointer(bond->primary_slave, slave); 3925 } else { /* we didn't change primary - exit */ 3926 break; 3927 } 3928 3929 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3930 primary ? slave_dev->name : "none"); 3931 3932 block_netpoll_tx(); 3933 bond_select_active_slave(bond); 3934 unblock_netpoll_tx(); 3935 break; 3936 case NETDEV_FEAT_CHANGE: 3937 if (!bond->notifier_ctx) { 3938 bond->notifier_ctx = true; 3939 netdev_compute_master_upper_features(bond->dev, true); 3940 bond->notifier_ctx = false; 3941 } 3942 break; 3943 case NETDEV_RESEND_IGMP: 3944 /* Propagate to master device */ 3945 call_netdevice_notifiers(event, slave->bond->dev); 3946 break; 3947 case NETDEV_XDP_FEAT_CHANGE: 3948 bond_xdp_set_features(bond_dev); 3949 break; 3950 default: 3951 break; 3952 } 3953 3954 return NOTIFY_DONE; 3955 } 3956 3957 /* bond_netdev_event: handle netdev notifier chain events. 3958 * 3959 * This function receives events for the netdev chain. The caller (an 3960 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3961 * locks for us to safely manipulate the slave devices (RTNL lock, 3962 * dev_probe_lock). 3963 */ 3964 static int bond_netdev_event(struct notifier_block *this, 3965 unsigned long event, void *ptr) 3966 { 3967 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3968 3969 netdev_dbg(event_dev, "%s received %s\n", 3970 __func__, netdev_cmd_to_name(event)); 3971 3972 if (!(event_dev->priv_flags & IFF_BONDING)) 3973 return NOTIFY_DONE; 3974 3975 if (event_dev->flags & IFF_MASTER) { 3976 int ret; 3977 3978 ret = bond_master_netdev_event(event, event_dev); 3979 if (ret != NOTIFY_DONE) 3980 return ret; 3981 } 3982 3983 if (event_dev->flags & IFF_SLAVE) 3984 return bond_slave_netdev_event(event, event_dev); 3985 3986 return NOTIFY_DONE; 3987 } 3988 3989 static struct notifier_block bond_netdev_notifier = { 3990 .notifier_call = bond_netdev_event, 3991 }; 3992 3993 /*---------------------------- Hashing Policies -----------------------------*/ 3994 3995 /* Helper to access data in a packet, with or without a backing skb. 3996 * If skb is given the data is linearized if necessary via pskb_may_pull. 3997 */ 3998 static inline const void *bond_pull_data(struct sk_buff *skb, 3999 const void *data, int hlen, int n) 4000 { 4001 if (likely(n <= hlen)) 4002 return data; 4003 else if (skb && likely(pskb_may_pull(skb, n))) 4004 return skb->data; 4005 4006 return NULL; 4007 } 4008 4009 /* L2 hash helper */ 4010 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4011 { 4012 struct ethhdr *ep; 4013 4014 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4015 if (!data) 4016 return 0; 4017 4018 ep = (struct ethhdr *)(data + mhoff); 4019 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 4020 } 4021 4022 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 4023 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 4024 { 4025 const struct ipv6hdr *iph6; 4026 const struct iphdr *iph; 4027 4028 if (l2_proto == htons(ETH_P_IP)) { 4029 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 4030 if (!data) 4031 return false; 4032 4033 iph = (const struct iphdr *)(data + *nhoff); 4034 iph_to_flow_copy_v4addrs(fk, iph); 4035 *nhoff += iph->ihl << 2; 4036 if (!ip_is_fragment(iph)) 4037 *ip_proto = iph->protocol; 4038 } else if (l2_proto == htons(ETH_P_IPV6)) { 4039 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 4040 if (!data) 4041 return false; 4042 4043 iph6 = (const struct ipv6hdr *)(data + *nhoff); 4044 iph_to_flow_copy_v6addrs(fk, iph6); 4045 *nhoff += sizeof(*iph6); 4046 *ip_proto = iph6->nexthdr; 4047 } else { 4048 return false; 4049 } 4050 4051 if (l34 && *ip_proto >= 0) 4052 fk->ports.ports = skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4053 4054 return true; 4055 } 4056 4057 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4058 { 4059 u32 srcmac_vendor = 0, srcmac_dev = 0; 4060 struct ethhdr *mac_hdr; 4061 u16 vlan = 0; 4062 int i; 4063 4064 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4065 if (!data) 4066 return 0; 4067 mac_hdr = (struct ethhdr *)(data + mhoff); 4068 4069 for (i = 0; i < 3; i++) 4070 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4071 4072 for (i = 3; i < ETH_ALEN; i++) 4073 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4074 4075 if (skb && skb_vlan_tag_present(skb)) 4076 vlan = skb_vlan_tag_get(skb); 4077 4078 return vlan ^ srcmac_vendor ^ srcmac_dev; 4079 } 4080 4081 /* Extract the appropriate headers based on bond's xmit policy */ 4082 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4083 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4084 { 4085 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4086 int ip_proto = -1; 4087 4088 switch (bond->params.xmit_policy) { 4089 case BOND_XMIT_POLICY_ENCAP23: 4090 case BOND_XMIT_POLICY_ENCAP34: 4091 memset(fk, 0, sizeof(*fk)); 4092 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4093 fk, data, l2_proto, nhoff, hlen, 0); 4094 default: 4095 break; 4096 } 4097 4098 fk->ports.ports = 0; 4099 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4100 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4101 return false; 4102 4103 /* ICMP error packets contains at least 8 bytes of the header 4104 * of the packet which generated the error. Use this information 4105 * to correlate ICMP error packets within the same flow which 4106 * generated the error. 4107 */ 4108 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4109 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4110 if (ip_proto == IPPROTO_ICMP) { 4111 if (!icmp_is_err(fk->icmp.type)) 4112 return true; 4113 4114 nhoff += sizeof(struct icmphdr); 4115 } else if (ip_proto == IPPROTO_ICMPV6) { 4116 if (!icmpv6_is_err(fk->icmp.type)) 4117 return true; 4118 4119 nhoff += sizeof(struct icmp6hdr); 4120 } 4121 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4122 } 4123 4124 return true; 4125 } 4126 4127 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4128 { 4129 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4130 (__force u32)flow_get_u32_src(flow); 4131 hash ^= (hash >> 16); 4132 hash ^= (hash >> 8); 4133 4134 /* discard lowest hash bit to deal with the common even ports pattern */ 4135 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4136 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4137 return hash >> 1; 4138 4139 return hash; 4140 } 4141 4142 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4143 * the data as required, but this function can be used without it if the data is 4144 * known to be linear (e.g. with xdp_buff). 4145 */ 4146 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4147 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4148 { 4149 struct flow_keys flow; 4150 u32 hash; 4151 4152 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4153 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4154 4155 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4156 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4157 return bond_eth_hash(skb, data, mhoff, hlen); 4158 4159 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4160 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4161 hash = bond_eth_hash(skb, data, mhoff, hlen); 4162 } else { 4163 if (flow.icmp.id) 4164 memcpy(&hash, &flow.icmp, sizeof(hash)); 4165 else 4166 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4167 } 4168 4169 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4170 } 4171 4172 /** 4173 * bond_xmit_hash - generate a hash value based on the xmit policy 4174 * @bond: bonding device 4175 * @skb: buffer to use for headers 4176 * 4177 * This function will extract the necessary headers from the skb buffer and use 4178 * them to generate a hash based on the xmit_policy set in the bonding device 4179 */ 4180 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4181 { 4182 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4183 skb->l4_hash) 4184 return skb->hash; 4185 4186 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4187 0, skb_network_offset(skb), 4188 skb_headlen(skb)); 4189 } 4190 4191 /** 4192 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4193 * @bond: bonding device 4194 * @xdp: buffer to use for headers 4195 * 4196 * The XDP variant of bond_xmit_hash. 4197 */ 4198 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4199 { 4200 struct ethhdr *eth; 4201 4202 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4203 return 0; 4204 4205 eth = (struct ethhdr *)xdp->data; 4206 4207 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4208 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4209 } 4210 4211 /*-------------------------- Device entry points ----------------------------*/ 4212 4213 void bond_work_init_all(struct bonding *bond) 4214 { 4215 INIT_DELAYED_WORK(&bond->mcast_work, 4216 bond_resend_igmp_join_requests_delayed); 4217 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4218 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4219 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4220 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4221 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4222 } 4223 4224 void bond_work_cancel_all(struct bonding *bond) 4225 { 4226 cancel_delayed_work_sync(&bond->mii_work); 4227 cancel_delayed_work_sync(&bond->arp_work); 4228 cancel_delayed_work_sync(&bond->alb_work); 4229 cancel_delayed_work_sync(&bond->ad_work); 4230 cancel_delayed_work_sync(&bond->mcast_work); 4231 cancel_delayed_work_sync(&bond->slave_arr_work); 4232 } 4233 4234 static int bond_open(struct net_device *bond_dev) 4235 { 4236 struct bonding *bond = netdev_priv(bond_dev); 4237 struct list_head *iter; 4238 struct slave *slave; 4239 4240 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4241 bond->rr_tx_counter = alloc_percpu(u32); 4242 if (!bond->rr_tx_counter) 4243 return -ENOMEM; 4244 } 4245 4246 /* reset slave->backup and slave->inactive */ 4247 if (bond_has_slaves(bond)) { 4248 bond_for_each_slave(bond, slave, iter) { 4249 if (bond_uses_primary(bond) && 4250 slave != rcu_access_pointer(bond->curr_active_slave)) { 4251 bond_set_slave_inactive_flags(slave, 4252 BOND_SLAVE_NOTIFY_NOW); 4253 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4254 bond_set_slave_active_flags(slave, 4255 BOND_SLAVE_NOTIFY_NOW); 4256 } 4257 } 4258 } 4259 4260 if (bond_is_lb(bond)) { 4261 /* bond_alb_initialize must be called before the timer 4262 * is started. 4263 */ 4264 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4265 return -ENOMEM; 4266 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4267 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4268 } 4269 4270 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4271 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4272 4273 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4274 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4275 bond->recv_probe = bond_rcv_validate; 4276 } 4277 4278 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4279 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4280 /* register to receive LACPDUs */ 4281 bond->recv_probe = bond_3ad_lacpdu_recv; 4282 bond_3ad_initiate_agg_selection(bond, 1); 4283 4284 bond_for_each_slave(bond, slave, iter) 4285 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4286 4287 if (bond->params.broadcast_neighbor) 4288 static_branch_inc(&bond_bcast_neigh_enabled); 4289 } 4290 4291 if (bond_mode_can_use_xmit_hash(bond)) 4292 bond_update_slave_arr(bond, NULL); 4293 4294 return 0; 4295 } 4296 4297 static int bond_close(struct net_device *bond_dev) 4298 { 4299 struct bonding *bond = netdev_priv(bond_dev); 4300 struct slave *slave; 4301 4302 bond_work_cancel_all(bond); 4303 bond->send_peer_notif = 0; 4304 if (bond_is_lb(bond)) 4305 bond_alb_deinitialize(bond); 4306 bond->recv_probe = NULL; 4307 4308 if (BOND_MODE(bond) == BOND_MODE_8023AD && 4309 bond->params.broadcast_neighbor) 4310 static_branch_dec(&bond_bcast_neigh_enabled); 4311 4312 if (bond_uses_primary(bond)) { 4313 rcu_read_lock(); 4314 slave = rcu_dereference(bond->curr_active_slave); 4315 if (slave) 4316 bond_hw_addr_flush(bond_dev, slave->dev); 4317 rcu_read_unlock(); 4318 } else { 4319 struct list_head *iter; 4320 4321 bond_for_each_slave(bond, slave, iter) 4322 bond_hw_addr_flush(bond_dev, slave->dev); 4323 } 4324 4325 return 0; 4326 } 4327 4328 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4329 * that some drivers can provide 32bit values only. 4330 */ 4331 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4332 const struct rtnl_link_stats64 *_new, 4333 const struct rtnl_link_stats64 *_old) 4334 { 4335 const u64 *new = (const u64 *)_new; 4336 const u64 *old = (const u64 *)_old; 4337 u64 *res = (u64 *)_res; 4338 int i; 4339 4340 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4341 u64 nv = new[i]; 4342 u64 ov = old[i]; 4343 s64 delta = nv - ov; 4344 4345 /* detects if this particular field is 32bit only */ 4346 if (((nv | ov) >> 32) == 0) 4347 delta = (s64)(s32)((u32)nv - (u32)ov); 4348 4349 /* filter anomalies, some drivers reset their stats 4350 * at down/up events. 4351 */ 4352 if (delta > 0) 4353 res[i] += delta; 4354 } 4355 } 4356 4357 #ifdef CONFIG_LOCKDEP 4358 static int bond_get_lowest_level_rcu(struct net_device *dev) 4359 { 4360 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4361 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4362 int cur = 0, max = 0; 4363 4364 now = dev; 4365 iter = &dev->adj_list.lower; 4366 4367 while (1) { 4368 next = NULL; 4369 while (1) { 4370 ldev = netdev_next_lower_dev_rcu(now, &iter); 4371 if (!ldev) 4372 break; 4373 4374 next = ldev; 4375 niter = &ldev->adj_list.lower; 4376 dev_stack[cur] = now; 4377 iter_stack[cur++] = iter; 4378 if (max <= cur) 4379 max = cur; 4380 break; 4381 } 4382 4383 if (!next) { 4384 if (!cur) 4385 return max; 4386 next = dev_stack[--cur]; 4387 niter = iter_stack[cur]; 4388 } 4389 4390 now = next; 4391 iter = niter; 4392 } 4393 4394 return max; 4395 } 4396 #endif 4397 4398 static void bond_get_stats(struct net_device *bond_dev, 4399 struct rtnl_link_stats64 *stats) 4400 { 4401 struct bonding *bond = netdev_priv(bond_dev); 4402 struct rtnl_link_stats64 temp; 4403 struct list_head *iter; 4404 struct slave *slave; 4405 int nest_level = 0; 4406 4407 4408 rcu_read_lock(); 4409 #ifdef CONFIG_LOCKDEP 4410 nest_level = bond_get_lowest_level_rcu(bond_dev); 4411 #endif 4412 4413 spin_lock_nested(&bond->stats_lock, nest_level); 4414 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4415 4416 bond_for_each_slave_rcu(bond, slave, iter) { 4417 const struct rtnl_link_stats64 *new = 4418 dev_get_stats(slave->dev, &temp); 4419 4420 bond_fold_stats(stats, new, &slave->slave_stats); 4421 4422 /* save off the slave stats for the next run */ 4423 memcpy(&slave->slave_stats, new, sizeof(*new)); 4424 } 4425 4426 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4427 spin_unlock(&bond->stats_lock); 4428 rcu_read_unlock(); 4429 } 4430 4431 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4432 { 4433 struct bonding *bond = netdev_priv(bond_dev); 4434 struct mii_ioctl_data *mii = NULL; 4435 4436 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4437 4438 switch (cmd) { 4439 case SIOCGMIIPHY: 4440 mii = if_mii(ifr); 4441 if (!mii) 4442 return -EINVAL; 4443 4444 mii->phy_id = 0; 4445 fallthrough; 4446 case SIOCGMIIREG: 4447 /* We do this again just in case we were called by SIOCGMIIREG 4448 * instead of SIOCGMIIPHY. 4449 */ 4450 mii = if_mii(ifr); 4451 if (!mii) 4452 return -EINVAL; 4453 4454 if (mii->reg_num == 1) { 4455 mii->val_out = 0; 4456 if (netif_carrier_ok(bond->dev)) 4457 mii->val_out = BMSR_LSTATUS; 4458 } 4459 4460 break; 4461 default: 4462 return -EOPNOTSUPP; 4463 } 4464 4465 return 0; 4466 } 4467 4468 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4469 { 4470 struct bonding *bond = netdev_priv(bond_dev); 4471 struct net_device *slave_dev = NULL; 4472 struct ifbond k_binfo; 4473 struct ifbond __user *u_binfo = NULL; 4474 struct ifslave k_sinfo; 4475 struct ifslave __user *u_sinfo = NULL; 4476 struct bond_opt_value newval; 4477 struct net *net; 4478 int res = 0; 4479 4480 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4481 4482 switch (cmd) { 4483 case SIOCBONDINFOQUERY: 4484 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4485 4486 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4487 return -EFAULT; 4488 4489 bond_info_query(bond_dev, &k_binfo); 4490 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4491 return -EFAULT; 4492 4493 return 0; 4494 case SIOCBONDSLAVEINFOQUERY: 4495 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4496 4497 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4498 return -EFAULT; 4499 4500 res = bond_slave_info_query(bond_dev, &k_sinfo); 4501 if (res == 0 && 4502 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4503 return -EFAULT; 4504 4505 return res; 4506 default: 4507 break; 4508 } 4509 4510 net = dev_net(bond_dev); 4511 4512 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4513 return -EPERM; 4514 4515 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4516 4517 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4518 4519 if (!slave_dev) 4520 return -ENODEV; 4521 4522 switch (cmd) { 4523 case SIOCBONDENSLAVE: 4524 res = bond_enslave(bond_dev, slave_dev, NULL); 4525 break; 4526 case SIOCBONDRELEASE: 4527 res = bond_release(bond_dev, slave_dev); 4528 break; 4529 case SIOCBONDSETHWADDR: 4530 res = bond_set_dev_addr(bond_dev, slave_dev); 4531 break; 4532 case SIOCBONDCHANGEACTIVE: 4533 bond_opt_initstr(&newval, slave_dev->name); 4534 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4535 &newval); 4536 break; 4537 default: 4538 res = -EOPNOTSUPP; 4539 } 4540 4541 return res; 4542 } 4543 4544 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4545 void __user *data, int cmd) 4546 { 4547 struct ifreq ifrdata = { .ifr_data = data }; 4548 4549 switch (cmd) { 4550 case BOND_INFO_QUERY_OLD: 4551 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4552 case BOND_SLAVE_INFO_QUERY_OLD: 4553 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4554 case BOND_ENSLAVE_OLD: 4555 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4556 case BOND_RELEASE_OLD: 4557 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4558 case BOND_SETHWADDR_OLD: 4559 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4560 case BOND_CHANGE_ACTIVE_OLD: 4561 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4562 } 4563 4564 return -EOPNOTSUPP; 4565 } 4566 4567 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4568 { 4569 struct bonding *bond = netdev_priv(bond_dev); 4570 4571 if (change & IFF_PROMISC) 4572 bond_set_promiscuity(bond, 4573 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4574 4575 if (change & IFF_ALLMULTI) 4576 bond_set_allmulti(bond, 4577 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4578 } 4579 4580 static void bond_set_rx_mode(struct net_device *bond_dev) 4581 { 4582 struct bonding *bond = netdev_priv(bond_dev); 4583 struct list_head *iter; 4584 struct slave *slave; 4585 4586 rcu_read_lock(); 4587 if (bond_uses_primary(bond)) { 4588 slave = rcu_dereference(bond->curr_active_slave); 4589 if (slave) { 4590 dev_uc_sync(slave->dev, bond_dev); 4591 dev_mc_sync(slave->dev, bond_dev); 4592 } 4593 } else { 4594 bond_for_each_slave_rcu(bond, slave, iter) { 4595 dev_uc_sync_multiple(slave->dev, bond_dev); 4596 dev_mc_sync_multiple(slave->dev, bond_dev); 4597 } 4598 } 4599 rcu_read_unlock(); 4600 } 4601 4602 static int bond_neigh_init(struct neighbour *n) 4603 { 4604 struct bonding *bond = netdev_priv(n->dev); 4605 const struct net_device_ops *slave_ops; 4606 struct neigh_parms parms; 4607 struct slave *slave; 4608 int ret = 0; 4609 4610 rcu_read_lock(); 4611 slave = bond_first_slave_rcu(bond); 4612 if (!slave) 4613 goto out; 4614 slave_ops = slave->dev->netdev_ops; 4615 if (!slave_ops->ndo_neigh_setup) 4616 goto out; 4617 4618 /* TODO: find another way [1] to implement this. 4619 * Passing a zeroed structure is fragile, 4620 * but at least we do not pass garbage. 4621 * 4622 * [1] One way would be that ndo_neigh_setup() never touch 4623 * struct neigh_parms, but propagate the new neigh_setup() 4624 * back to ___neigh_create() / neigh_parms_alloc() 4625 */ 4626 memset(&parms, 0, sizeof(parms)); 4627 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4628 4629 if (ret) 4630 goto out; 4631 4632 if (parms.neigh_setup) 4633 ret = parms.neigh_setup(n); 4634 out: 4635 rcu_read_unlock(); 4636 return ret; 4637 } 4638 4639 /* The bonding ndo_neigh_setup is called at init time beofre any 4640 * slave exists. So we must declare proxy setup function which will 4641 * be used at run time to resolve the actual slave neigh param setup. 4642 * 4643 * It's also called by master devices (such as vlans) to setup their 4644 * underlying devices. In that case - do nothing, we're already set up from 4645 * our init. 4646 */ 4647 static int bond_neigh_setup(struct net_device *dev, 4648 struct neigh_parms *parms) 4649 { 4650 /* modify only our neigh_parms */ 4651 if (parms->dev == dev) 4652 parms->neigh_setup = bond_neigh_init; 4653 4654 return 0; 4655 } 4656 4657 /* Change the MTU of all of a master's slaves to match the master */ 4658 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4659 { 4660 struct bonding *bond = netdev_priv(bond_dev); 4661 struct slave *slave, *rollback_slave; 4662 struct list_head *iter; 4663 int res = 0; 4664 4665 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4666 4667 bond_for_each_slave(bond, slave, iter) { 4668 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4669 slave, slave->dev->netdev_ops->ndo_change_mtu); 4670 4671 res = dev_set_mtu(slave->dev, new_mtu); 4672 4673 if (res) { 4674 /* If we failed to set the slave's mtu to the new value 4675 * we must abort the operation even in ACTIVE_BACKUP 4676 * mode, because if we allow the backup slaves to have 4677 * different mtu values than the active slave we'll 4678 * need to change their mtu when doing a failover. That 4679 * means changing their mtu from timer context, which 4680 * is probably not a good idea. 4681 */ 4682 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4683 res, new_mtu); 4684 goto unwind; 4685 } 4686 } 4687 4688 WRITE_ONCE(bond_dev->mtu, new_mtu); 4689 4690 return 0; 4691 4692 unwind: 4693 /* unwind from head to the slave that failed */ 4694 bond_for_each_slave(bond, rollback_slave, iter) { 4695 int tmp_res; 4696 4697 if (rollback_slave == slave) 4698 break; 4699 4700 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4701 if (tmp_res) 4702 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4703 tmp_res); 4704 } 4705 4706 return res; 4707 } 4708 4709 /* Change HW address 4710 * 4711 * Note that many devices must be down to change the HW address, and 4712 * downing the master releases all slaves. We can make bonds full of 4713 * bonding devices to test this, however. 4714 */ 4715 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4716 { 4717 struct bonding *bond = netdev_priv(bond_dev); 4718 struct slave *slave, *rollback_slave; 4719 struct sockaddr_storage *ss = addr, tmp_ss; 4720 struct list_head *iter; 4721 int res = 0; 4722 4723 if (BOND_MODE(bond) == BOND_MODE_ALB) 4724 return bond_alb_set_mac_address(bond_dev, addr); 4725 4726 4727 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4728 4729 /* If fail_over_mac is enabled, do nothing and return success. 4730 * Returning an error causes ifenslave to fail. 4731 */ 4732 if (bond->params.fail_over_mac && 4733 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4734 return 0; 4735 4736 if (!is_valid_ether_addr(ss->__data)) 4737 return -EADDRNOTAVAIL; 4738 4739 bond_for_each_slave(bond, slave, iter) { 4740 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4741 __func__, slave); 4742 res = dev_set_mac_address(slave->dev, addr, NULL); 4743 if (res) { 4744 /* TODO: consider downing the slave 4745 * and retry ? 4746 * User should expect communications 4747 * breakage anyway until ARP finish 4748 * updating, so... 4749 */ 4750 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4751 __func__, res); 4752 goto unwind; 4753 } 4754 } 4755 4756 /* success */ 4757 dev_addr_set(bond_dev, ss->__data); 4758 return 0; 4759 4760 unwind: 4761 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4762 tmp_ss.ss_family = bond_dev->type; 4763 4764 /* unwind from head to the slave that failed */ 4765 bond_for_each_slave(bond, rollback_slave, iter) { 4766 int tmp_res; 4767 4768 if (rollback_slave == slave) 4769 break; 4770 4771 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_ss, NULL); 4772 if (tmp_res) { 4773 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4774 __func__, tmp_res); 4775 } 4776 } 4777 4778 return res; 4779 } 4780 4781 /** 4782 * bond_get_slave_by_id - get xmit slave with slave_id 4783 * @bond: bonding device that is transmitting 4784 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4785 * 4786 * This function tries to get slave with slave_id but in case 4787 * it fails, it tries to find the first available slave for transmission. 4788 */ 4789 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4790 int slave_id) 4791 { 4792 struct list_head *iter; 4793 struct slave *slave; 4794 int i = slave_id; 4795 4796 /* Here we start from the slave with slave_id */ 4797 bond_for_each_slave_rcu(bond, slave, iter) { 4798 if (--i < 0) { 4799 if (bond_slave_can_tx(slave)) 4800 return slave; 4801 } 4802 } 4803 4804 /* Here we start from the first slave up to slave_id */ 4805 i = slave_id; 4806 bond_for_each_slave_rcu(bond, slave, iter) { 4807 if (--i < 0) 4808 break; 4809 if (bond_slave_can_tx(slave)) 4810 return slave; 4811 } 4812 /* no slave that can tx has been found */ 4813 return NULL; 4814 } 4815 4816 /** 4817 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4818 * @bond: bonding device to use 4819 * 4820 * Based on the value of the bonding device's packets_per_slave parameter 4821 * this function generates a slave id, which is usually used as the next 4822 * slave to transmit through. 4823 */ 4824 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4825 { 4826 u32 slave_id; 4827 struct reciprocal_value reciprocal_packets_per_slave; 4828 int packets_per_slave = bond->params.packets_per_slave; 4829 4830 switch (packets_per_slave) { 4831 case 0: 4832 slave_id = get_random_u32(); 4833 break; 4834 case 1: 4835 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4836 break; 4837 default: 4838 reciprocal_packets_per_slave = 4839 bond->params.reciprocal_packets_per_slave; 4840 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4841 slave_id = reciprocal_divide(slave_id, 4842 reciprocal_packets_per_slave); 4843 break; 4844 } 4845 4846 return slave_id; 4847 } 4848 4849 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4850 struct sk_buff *skb) 4851 { 4852 struct slave *slave; 4853 int slave_cnt; 4854 u32 slave_id; 4855 4856 /* Start with the curr_active_slave that joined the bond as the 4857 * default for sending IGMP traffic. For failover purposes one 4858 * needs to maintain some consistency for the interface that will 4859 * send the join/membership reports. The curr_active_slave found 4860 * will send all of this type of traffic. 4861 */ 4862 if (skb->protocol == htons(ETH_P_IP)) { 4863 int noff = skb_network_offset(skb); 4864 struct iphdr *iph; 4865 4866 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4867 goto non_igmp; 4868 4869 iph = ip_hdr(skb); 4870 if (iph->protocol == IPPROTO_IGMP) { 4871 slave = rcu_dereference(bond->curr_active_slave); 4872 if (slave) 4873 return slave; 4874 return bond_get_slave_by_id(bond, 0); 4875 } 4876 } 4877 4878 non_igmp: 4879 slave_cnt = READ_ONCE(bond->slave_cnt); 4880 if (likely(slave_cnt)) { 4881 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4882 return bond_get_slave_by_id(bond, slave_id); 4883 } 4884 return NULL; 4885 } 4886 4887 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4888 struct xdp_buff *xdp) 4889 { 4890 struct slave *slave; 4891 int slave_cnt; 4892 u32 slave_id; 4893 const struct ethhdr *eth; 4894 void *data = xdp->data; 4895 4896 if (data + sizeof(struct ethhdr) > xdp->data_end) 4897 goto non_igmp; 4898 4899 eth = (struct ethhdr *)data; 4900 data += sizeof(struct ethhdr); 4901 4902 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4903 if (eth->h_proto == htons(ETH_P_IP)) { 4904 const struct iphdr *iph; 4905 4906 if (data + sizeof(struct iphdr) > xdp->data_end) 4907 goto non_igmp; 4908 4909 iph = (struct iphdr *)data; 4910 4911 if (iph->protocol == IPPROTO_IGMP) { 4912 slave = rcu_dereference(bond->curr_active_slave); 4913 if (slave) 4914 return slave; 4915 return bond_get_slave_by_id(bond, 0); 4916 } 4917 } 4918 4919 non_igmp: 4920 slave_cnt = READ_ONCE(bond->slave_cnt); 4921 if (likely(slave_cnt)) { 4922 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4923 return bond_get_slave_by_id(bond, slave_id); 4924 } 4925 return NULL; 4926 } 4927 4928 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4929 struct net_device *bond_dev) 4930 { 4931 struct bonding *bond = netdev_priv(bond_dev); 4932 struct slave *slave; 4933 4934 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4935 if (likely(slave)) 4936 return bond_dev_queue_xmit(bond, skb, slave->dev); 4937 4938 return bond_tx_drop(bond_dev, skb); 4939 } 4940 4941 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 4942 { 4943 return rcu_dereference(bond->curr_active_slave); 4944 } 4945 4946 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4947 * the bond has a usable interface. 4948 */ 4949 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4950 struct net_device *bond_dev) 4951 { 4952 struct bonding *bond = netdev_priv(bond_dev); 4953 struct slave *slave; 4954 4955 slave = bond_xmit_activebackup_slave_get(bond); 4956 if (slave) 4957 return bond_dev_queue_xmit(bond, skb, slave->dev); 4958 4959 return bond_tx_drop(bond_dev, skb); 4960 } 4961 4962 /* Use this to update slave_array when (a) it's not appropriate to update 4963 * slave_array right away (note that update_slave_array() may sleep) 4964 * and / or (b) RTNL is not held. 4965 */ 4966 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4967 { 4968 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4969 } 4970 4971 /* Slave array work handler. Holds only RTNL */ 4972 static void bond_slave_arr_handler(struct work_struct *work) 4973 { 4974 struct bonding *bond = container_of(work, struct bonding, 4975 slave_arr_work.work); 4976 int ret; 4977 4978 if (!rtnl_trylock()) 4979 goto err; 4980 4981 ret = bond_update_slave_arr(bond, NULL); 4982 rtnl_unlock(); 4983 if (ret) { 4984 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4985 goto err; 4986 } 4987 return; 4988 4989 err: 4990 bond_slave_arr_work_rearm(bond, 1); 4991 } 4992 4993 static void bond_skip_slave(struct bond_up_slave *slaves, 4994 struct slave *skipslave) 4995 { 4996 int idx; 4997 4998 /* Rare situation where caller has asked to skip a specific 4999 * slave but allocation failed (most likely!). BTW this is 5000 * only possible when the call is initiated from 5001 * __bond_release_one(). In this situation; overwrite the 5002 * skipslave entry in the array with the last entry from the 5003 * array to avoid a situation where the xmit path may choose 5004 * this to-be-skipped slave to send a packet out. 5005 */ 5006 for (idx = 0; slaves && idx < slaves->count; idx++) { 5007 if (skipslave == slaves->arr[idx]) { 5008 slaves->arr[idx] = 5009 slaves->arr[slaves->count - 1]; 5010 slaves->count--; 5011 break; 5012 } 5013 } 5014 } 5015 5016 static void bond_set_slave_arr(struct bonding *bond, 5017 struct bond_up_slave *usable_slaves, 5018 struct bond_up_slave *all_slaves) 5019 { 5020 struct bond_up_slave *usable, *all; 5021 5022 usable = rtnl_dereference(bond->usable_slaves); 5023 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5024 kfree_rcu(usable, rcu); 5025 5026 all = rtnl_dereference(bond->all_slaves); 5027 rcu_assign_pointer(bond->all_slaves, all_slaves); 5028 kfree_rcu(all, rcu); 5029 } 5030 5031 static void bond_reset_slave_arr(struct bonding *bond) 5032 { 5033 bond_set_slave_arr(bond, NULL, NULL); 5034 } 5035 5036 /* Build the usable slaves array in control path for modes that use xmit-hash 5037 * to determine the slave interface - 5038 * (a) BOND_MODE_8023AD 5039 * (b) BOND_MODE_XOR 5040 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5041 * 5042 * The caller is expected to hold RTNL only and NO other lock! 5043 */ 5044 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5045 { 5046 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5047 struct slave *slave; 5048 struct list_head *iter; 5049 int agg_id = 0; 5050 int ret = 0; 5051 5052 might_sleep(); 5053 5054 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5055 bond->slave_cnt), GFP_KERNEL); 5056 all_slaves = kzalloc(struct_size(all_slaves, arr, 5057 bond->slave_cnt), GFP_KERNEL); 5058 if (!usable_slaves || !all_slaves) { 5059 ret = -ENOMEM; 5060 goto out; 5061 } 5062 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5063 struct ad_info ad_info; 5064 5065 spin_lock_bh(&bond->mode_lock); 5066 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5067 spin_unlock_bh(&bond->mode_lock); 5068 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5069 /* No active aggragator means it's not safe to use 5070 * the previous array. 5071 */ 5072 bond_reset_slave_arr(bond); 5073 goto out; 5074 } 5075 spin_unlock_bh(&bond->mode_lock); 5076 agg_id = ad_info.aggregator_id; 5077 } 5078 bond_for_each_slave(bond, slave, iter) { 5079 if (skipslave == slave) 5080 continue; 5081 5082 all_slaves->arr[all_slaves->count++] = slave; 5083 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5084 struct aggregator *agg; 5085 5086 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5087 if (!agg || agg->aggregator_identifier != agg_id) 5088 continue; 5089 } 5090 if (!bond_slave_can_tx(slave)) 5091 continue; 5092 5093 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5094 usable_slaves->count); 5095 5096 usable_slaves->arr[usable_slaves->count++] = slave; 5097 } 5098 5099 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5100 return ret; 5101 out: 5102 if (ret != 0 && skipslave) { 5103 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5104 skipslave); 5105 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5106 skipslave); 5107 } 5108 kfree_rcu(all_slaves, rcu); 5109 kfree_rcu(usable_slaves, rcu); 5110 5111 return ret; 5112 } 5113 5114 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5115 struct sk_buff *skb, 5116 struct bond_up_slave *slaves) 5117 { 5118 struct slave *slave; 5119 unsigned int count; 5120 u32 hash; 5121 5122 hash = bond_xmit_hash(bond, skb); 5123 count = slaves ? READ_ONCE(slaves->count) : 0; 5124 if (unlikely(!count)) 5125 return NULL; 5126 5127 slave = slaves->arr[hash % count]; 5128 return slave; 5129 } 5130 5131 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5132 struct xdp_buff *xdp) 5133 { 5134 struct bond_up_slave *slaves; 5135 unsigned int count; 5136 u32 hash; 5137 5138 hash = bond_xmit_hash_xdp(bond, xdp); 5139 slaves = rcu_dereference(bond->usable_slaves); 5140 count = slaves ? READ_ONCE(slaves->count) : 0; 5141 if (unlikely(!count)) 5142 return NULL; 5143 5144 return slaves->arr[hash % count]; 5145 } 5146 5147 static bool bond_should_broadcast_neighbor(struct sk_buff *skb, 5148 struct net_device *dev) 5149 { 5150 struct bonding *bond = netdev_priv(dev); 5151 struct { 5152 struct ipv6hdr ip6; 5153 struct icmp6hdr icmp6; 5154 } *combined, _combined; 5155 5156 if (!static_branch_unlikely(&bond_bcast_neigh_enabled)) 5157 return false; 5158 5159 if (!bond->params.broadcast_neighbor) 5160 return false; 5161 5162 if (skb->protocol == htons(ETH_P_ARP)) 5163 return true; 5164 5165 if (skb->protocol == htons(ETH_P_IPV6)) { 5166 combined = skb_header_pointer(skb, skb_mac_header_len(skb), 5167 sizeof(_combined), 5168 &_combined); 5169 if (combined && combined->ip6.nexthdr == NEXTHDR_ICMP && 5170 (combined->icmp6.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION || 5171 combined->icmp6.icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) 5172 return true; 5173 } 5174 5175 return false; 5176 } 5177 5178 /* Use this Xmit function for 3AD as well as XOR modes. The current 5179 * usable slave array is formed in the control path. The xmit function 5180 * just calculates hash and sends the packet out. 5181 */ 5182 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5183 struct net_device *dev) 5184 { 5185 struct bonding *bond = netdev_priv(dev); 5186 struct bond_up_slave *slaves; 5187 struct slave *slave; 5188 5189 slaves = rcu_dereference(bond->usable_slaves); 5190 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5191 if (likely(slave)) 5192 return bond_dev_queue_xmit(bond, skb, slave->dev); 5193 5194 return bond_tx_drop(dev, skb); 5195 } 5196 5197 /* in broadcast mode, we send everything to all or usable slave interfaces. 5198 * under rcu_read_lock when this function is called. 5199 */ 5200 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5201 struct net_device *bond_dev, 5202 bool all_slaves) 5203 { 5204 struct bonding *bond = netdev_priv(bond_dev); 5205 struct bond_up_slave *slaves; 5206 bool xmit_suc = false; 5207 bool skb_used = false; 5208 int slaves_count, i; 5209 5210 if (all_slaves) 5211 slaves = rcu_dereference(bond->all_slaves); 5212 else 5213 slaves = rcu_dereference(bond->usable_slaves); 5214 5215 slaves_count = slaves ? READ_ONCE(slaves->count) : 0; 5216 for (i = 0; i < slaves_count; i++) { 5217 struct slave *slave = slaves->arr[i]; 5218 struct sk_buff *skb2; 5219 5220 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5221 continue; 5222 5223 if (bond_is_last_slave(bond, slave)) { 5224 skb2 = skb; 5225 skb_used = true; 5226 } else { 5227 skb2 = skb_clone(skb, GFP_ATOMIC); 5228 if (!skb2) { 5229 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5230 bond_dev->name, __func__); 5231 continue; 5232 } 5233 } 5234 5235 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5236 xmit_suc = true; 5237 } 5238 5239 if (!skb_used) 5240 dev_kfree_skb_any(skb); 5241 5242 if (xmit_suc) 5243 return NETDEV_TX_OK; 5244 5245 dev_core_stats_tx_dropped_inc(bond_dev); 5246 return NET_XMIT_DROP; 5247 } 5248 5249 /*------------------------- Device initialization ---------------------------*/ 5250 5251 /* Lookup the slave that corresponds to a qid */ 5252 static inline int bond_slave_override(struct bonding *bond, 5253 struct sk_buff *skb) 5254 { 5255 struct slave *slave = NULL; 5256 struct list_head *iter; 5257 5258 if (!skb_rx_queue_recorded(skb)) 5259 return 1; 5260 5261 /* Find out if any slaves have the same mapping as this skb. */ 5262 bond_for_each_slave_rcu(bond, slave, iter) { 5263 if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) { 5264 if (bond_slave_is_up(slave) && 5265 slave->link == BOND_LINK_UP) { 5266 bond_dev_queue_xmit(bond, skb, slave->dev); 5267 return 0; 5268 } 5269 /* If the slave isn't UP, use default transmit policy. */ 5270 break; 5271 } 5272 } 5273 5274 return 1; 5275 } 5276 5277 5278 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5279 struct net_device *sb_dev) 5280 { 5281 /* This helper function exists to help dev_pick_tx get the correct 5282 * destination queue. Using a helper function skips a call to 5283 * skb_tx_hash and will put the skbs in the queue we expect on their 5284 * way down to the bonding driver. 5285 */ 5286 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5287 5288 /* Save the original txq to restore before passing to the driver */ 5289 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5290 5291 if (unlikely(txq >= dev->real_num_tx_queues)) { 5292 do { 5293 txq -= dev->real_num_tx_queues; 5294 } while (txq >= dev->real_num_tx_queues); 5295 } 5296 return txq; 5297 } 5298 5299 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5300 struct sk_buff *skb, 5301 bool all_slaves) 5302 { 5303 struct bonding *bond = netdev_priv(master_dev); 5304 struct bond_up_slave *slaves; 5305 struct slave *slave = NULL; 5306 5307 switch (BOND_MODE(bond)) { 5308 case BOND_MODE_ROUNDROBIN: 5309 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5310 break; 5311 case BOND_MODE_ACTIVEBACKUP: 5312 slave = bond_xmit_activebackup_slave_get(bond); 5313 break; 5314 case BOND_MODE_8023AD: 5315 case BOND_MODE_XOR: 5316 if (all_slaves) 5317 slaves = rcu_dereference(bond->all_slaves); 5318 else 5319 slaves = rcu_dereference(bond->usable_slaves); 5320 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5321 break; 5322 case BOND_MODE_BROADCAST: 5323 break; 5324 case BOND_MODE_ALB: 5325 slave = bond_xmit_alb_slave_get(bond, skb); 5326 break; 5327 case BOND_MODE_TLB: 5328 slave = bond_xmit_tlb_slave_get(bond, skb); 5329 break; 5330 default: 5331 /* Should never happen, mode already checked */ 5332 WARN_ONCE(true, "Unknown bonding mode"); 5333 break; 5334 } 5335 5336 if (slave) 5337 return slave->dev; 5338 return NULL; 5339 } 5340 5341 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5342 { 5343 switch (sk->sk_family) { 5344 #if IS_ENABLED(CONFIG_IPV6) 5345 case AF_INET6: 5346 if (ipv6_only_sock(sk) || 5347 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5348 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5349 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5350 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5351 break; 5352 } 5353 fallthrough; 5354 #endif 5355 default: /* AF_INET */ 5356 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5357 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5358 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5359 break; 5360 } 5361 5362 flow->ports.src = inet_sk(sk)->inet_sport; 5363 flow->ports.dst = inet_sk(sk)->inet_dport; 5364 } 5365 5366 /** 5367 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5368 * @sk: socket to use for headers 5369 * 5370 * This function will extract the necessary field from the socket and use 5371 * them to generate a hash based on the LAYER34 xmit_policy. 5372 * Assumes that sk is a TCP or UDP socket. 5373 */ 5374 static u32 bond_sk_hash_l34(struct sock *sk) 5375 { 5376 struct flow_keys flow; 5377 u32 hash; 5378 5379 bond_sk_to_flow(sk, &flow); 5380 5381 /* L4 */ 5382 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5383 /* L3 */ 5384 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5385 } 5386 5387 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5388 struct sock *sk) 5389 { 5390 struct bond_up_slave *slaves; 5391 struct slave *slave; 5392 unsigned int count; 5393 u32 hash; 5394 5395 slaves = rcu_dereference(bond->usable_slaves); 5396 count = slaves ? READ_ONCE(slaves->count) : 0; 5397 if (unlikely(!count)) 5398 return NULL; 5399 5400 hash = bond_sk_hash_l34(sk); 5401 slave = slaves->arr[hash % count]; 5402 5403 return slave->dev; 5404 } 5405 5406 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5407 struct sock *sk) 5408 { 5409 struct bonding *bond = netdev_priv(dev); 5410 struct net_device *lower = NULL; 5411 5412 rcu_read_lock(); 5413 if (bond_sk_check(bond)) 5414 lower = __bond_sk_get_lower_dev(bond, sk); 5415 rcu_read_unlock(); 5416 5417 return lower; 5418 } 5419 5420 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5421 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5422 struct net_device *dev) 5423 { 5424 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5425 5426 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded 5427 * was true, if tls_device_down is running in parallel, but it's OK, 5428 * because bond_get_slave_by_dev has a NULL check. 5429 */ 5430 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5431 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5432 return bond_tx_drop(dev, skb); 5433 } 5434 #endif 5435 5436 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5437 { 5438 struct bonding *bond = netdev_priv(dev); 5439 5440 if (bond_should_override_tx_queue(bond) && 5441 !bond_slave_override(bond, skb)) 5442 return NETDEV_TX_OK; 5443 5444 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5445 if (tls_is_skb_tx_device_offloaded(skb)) 5446 return bond_tls_device_xmit(bond, skb, dev); 5447 #endif 5448 5449 switch (BOND_MODE(bond)) { 5450 case BOND_MODE_ROUNDROBIN: 5451 return bond_xmit_roundrobin(skb, dev); 5452 case BOND_MODE_ACTIVEBACKUP: 5453 return bond_xmit_activebackup(skb, dev); 5454 case BOND_MODE_8023AD: 5455 if (bond_should_broadcast_neighbor(skb, dev)) 5456 return bond_xmit_broadcast(skb, dev, false); 5457 fallthrough; 5458 case BOND_MODE_XOR: 5459 return bond_3ad_xor_xmit(skb, dev); 5460 case BOND_MODE_BROADCAST: 5461 return bond_xmit_broadcast(skb, dev, true); 5462 case BOND_MODE_ALB: 5463 return bond_alb_xmit(skb, dev); 5464 case BOND_MODE_TLB: 5465 return bond_tlb_xmit(skb, dev); 5466 default: 5467 /* Should never happen, mode already checked */ 5468 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5469 WARN_ON_ONCE(1); 5470 return bond_tx_drop(dev, skb); 5471 } 5472 } 5473 5474 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5475 { 5476 struct bonding *bond = netdev_priv(dev); 5477 netdev_tx_t ret = NETDEV_TX_OK; 5478 5479 /* If we risk deadlock from transmitting this in the 5480 * netpoll path, tell netpoll to queue the frame for later tx 5481 */ 5482 if (unlikely(is_netpoll_tx_blocked(dev))) 5483 return NETDEV_TX_BUSY; 5484 5485 rcu_read_lock(); 5486 if (bond_has_slaves(bond)) 5487 ret = __bond_start_xmit(skb, dev); 5488 else 5489 ret = bond_tx_drop(dev, skb); 5490 rcu_read_unlock(); 5491 5492 return ret; 5493 } 5494 5495 static struct net_device * 5496 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5497 { 5498 struct bonding *bond = netdev_priv(bond_dev); 5499 struct slave *slave; 5500 5501 /* Caller needs to hold rcu_read_lock() */ 5502 5503 switch (BOND_MODE(bond)) { 5504 case BOND_MODE_ROUNDROBIN: 5505 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5506 break; 5507 5508 case BOND_MODE_ACTIVEBACKUP: 5509 slave = bond_xmit_activebackup_slave_get(bond); 5510 break; 5511 5512 case BOND_MODE_8023AD: 5513 case BOND_MODE_XOR: 5514 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5515 break; 5516 5517 default: 5518 if (net_ratelimit()) 5519 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", 5520 BOND_MODE(bond)); 5521 return NULL; 5522 } 5523 5524 if (slave) 5525 return slave->dev; 5526 5527 return NULL; 5528 } 5529 5530 static int bond_xdp_xmit(struct net_device *bond_dev, 5531 int n, struct xdp_frame **frames, u32 flags) 5532 { 5533 int nxmit, err = -ENXIO; 5534 5535 rcu_read_lock(); 5536 5537 for (nxmit = 0; nxmit < n; nxmit++) { 5538 struct xdp_frame *frame = frames[nxmit]; 5539 struct xdp_frame *frames1[] = {frame}; 5540 struct net_device *slave_dev; 5541 struct xdp_buff xdp; 5542 5543 xdp_convert_frame_to_buff(frame, &xdp); 5544 5545 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5546 if (!slave_dev) { 5547 err = -ENXIO; 5548 break; 5549 } 5550 5551 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5552 if (err < 1) 5553 break; 5554 } 5555 5556 rcu_read_unlock(); 5557 5558 /* If error happened on the first frame then we can pass the error up, otherwise 5559 * report the number of frames that were xmitted. 5560 */ 5561 if (err < 0) 5562 return (nxmit == 0 ? err : nxmit); 5563 5564 return nxmit; 5565 } 5566 5567 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5568 struct netlink_ext_ack *extack) 5569 { 5570 struct bonding *bond = netdev_priv(dev); 5571 struct list_head *iter; 5572 struct slave *slave, *rollback_slave; 5573 struct bpf_prog *old_prog; 5574 struct netdev_bpf xdp = { 5575 .command = XDP_SETUP_PROG, 5576 .flags = 0, 5577 .prog = prog, 5578 .extack = extack, 5579 }; 5580 int err; 5581 5582 ASSERT_RTNL(); 5583 5584 if (!bond_xdp_check(bond, BOND_MODE(bond))) { 5585 BOND_NL_ERR(dev, extack, 5586 "No native XDP support for the current bonding mode"); 5587 return -EOPNOTSUPP; 5588 } 5589 5590 old_prog = bond->xdp_prog; 5591 bond->xdp_prog = prog; 5592 5593 bond_for_each_slave(bond, slave, iter) { 5594 struct net_device *slave_dev = slave->dev; 5595 5596 if (!slave_dev->netdev_ops->ndo_bpf || 5597 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5598 SLAVE_NL_ERR(dev, slave_dev, extack, 5599 "Slave device does not support XDP"); 5600 err = -EOPNOTSUPP; 5601 goto err; 5602 } 5603 5604 if (dev_xdp_prog_count(slave_dev) > 0) { 5605 SLAVE_NL_ERR(dev, slave_dev, extack, 5606 "Slave has XDP program loaded, please unload before enslaving"); 5607 err = -EOPNOTSUPP; 5608 goto err; 5609 } 5610 5611 err = dev_xdp_propagate(slave_dev, &xdp); 5612 if (err < 0) { 5613 /* ndo_bpf() sets extack error message */ 5614 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5615 goto err; 5616 } 5617 if (prog) 5618 bpf_prog_inc(prog); 5619 } 5620 5621 if (prog) { 5622 static_branch_inc(&bpf_master_redirect_enabled_key); 5623 } else if (old_prog) { 5624 bpf_prog_put(old_prog); 5625 static_branch_dec(&bpf_master_redirect_enabled_key); 5626 } 5627 5628 return 0; 5629 5630 err: 5631 /* unwind the program changes */ 5632 bond->xdp_prog = old_prog; 5633 xdp.prog = old_prog; 5634 xdp.extack = NULL; /* do not overwrite original error */ 5635 5636 bond_for_each_slave(bond, rollback_slave, iter) { 5637 struct net_device *slave_dev = rollback_slave->dev; 5638 int err_unwind; 5639 5640 if (slave == rollback_slave) 5641 break; 5642 5643 err_unwind = dev_xdp_propagate(slave_dev, &xdp); 5644 if (err_unwind < 0) 5645 slave_err(dev, slave_dev, 5646 "Error %d when unwinding XDP program change\n", err_unwind); 5647 else if (xdp.prog) 5648 bpf_prog_inc(xdp.prog); 5649 } 5650 return err; 5651 } 5652 5653 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5654 { 5655 switch (xdp->command) { 5656 case XDP_SETUP_PROG: 5657 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5658 default: 5659 return -EINVAL; 5660 } 5661 } 5662 5663 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5664 { 5665 if (speed == 0 || speed == SPEED_UNKNOWN) 5666 speed = slave->speed; 5667 else 5668 speed = min(speed, slave->speed); 5669 5670 return speed; 5671 } 5672 5673 /* Set the BOND_PHC_INDEX flag to notify user space */ 5674 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg) 5675 { 5676 struct ifreq *ifr = kernel_cfg->ifr; 5677 struct hwtstamp_config cfg; 5678 5679 if (kernel_cfg->copied_to_user) { 5680 /* Lower device has a legacy implementation */ 5681 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 5682 return -EFAULT; 5683 5684 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5685 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg))) 5686 return -EFAULT; 5687 } else { 5688 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5689 } 5690 5691 return 0; 5692 } 5693 5694 static int bond_hwtstamp_get(struct net_device *dev, 5695 struct kernel_hwtstamp_config *cfg) 5696 { 5697 struct bonding *bond = netdev_priv(dev); 5698 struct net_device *real_dev; 5699 int err; 5700 5701 real_dev = bond_option_active_slave_get_rcu(bond); 5702 if (!real_dev) 5703 return -EOPNOTSUPP; 5704 5705 err = generic_hwtstamp_get_lower(real_dev, cfg); 5706 if (err) 5707 return err; 5708 5709 return bond_set_phc_index_flag(cfg); 5710 } 5711 5712 static int bond_hwtstamp_set(struct net_device *dev, 5713 struct kernel_hwtstamp_config *cfg, 5714 struct netlink_ext_ack *extack) 5715 { 5716 struct bonding *bond = netdev_priv(dev); 5717 struct net_device *real_dev; 5718 int err; 5719 5720 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 5721 return -EOPNOTSUPP; 5722 5723 real_dev = bond_option_active_slave_get_rcu(bond); 5724 if (!real_dev) 5725 return -EOPNOTSUPP; 5726 5727 err = generic_hwtstamp_set_lower(real_dev, cfg, extack); 5728 if (err) 5729 return err; 5730 5731 return bond_set_phc_index_flag(cfg); 5732 } 5733 5734 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5735 struct ethtool_link_ksettings *cmd) 5736 { 5737 struct bonding *bond = netdev_priv(bond_dev); 5738 struct list_head *iter; 5739 struct slave *slave; 5740 u32 speed = 0; 5741 5742 cmd->base.duplex = DUPLEX_UNKNOWN; 5743 cmd->base.port = PORT_OTHER; 5744 5745 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5746 * do not need to check mode. Though link speed might not represent 5747 * the true receive or transmit bandwidth (not all modes are symmetric) 5748 * this is an accurate maximum. 5749 */ 5750 bond_for_each_slave(bond, slave, iter) { 5751 if (bond_slave_can_tx(slave)) { 5752 bond_update_speed_duplex(slave); 5753 if (slave->speed != SPEED_UNKNOWN) { 5754 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5755 speed = bond_mode_bcast_speed(slave, 5756 speed); 5757 else 5758 speed += slave->speed; 5759 } 5760 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5761 slave->duplex != DUPLEX_UNKNOWN) 5762 cmd->base.duplex = slave->duplex; 5763 } 5764 } 5765 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5766 5767 return 0; 5768 } 5769 5770 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5771 struct ethtool_drvinfo *drvinfo) 5772 { 5773 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5774 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5775 BOND_ABI_VERSION); 5776 } 5777 5778 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5779 struct kernel_ethtool_ts_info *info) 5780 { 5781 struct bonding *bond = netdev_priv(bond_dev); 5782 struct kernel_ethtool_ts_info ts_info; 5783 struct net_device *real_dev; 5784 bool sw_tx_support = false; 5785 struct list_head *iter; 5786 struct slave *slave; 5787 int ret = 0; 5788 5789 rcu_read_lock(); 5790 real_dev = bond_option_active_slave_get_rcu(bond); 5791 dev_hold(real_dev); 5792 rcu_read_unlock(); 5793 5794 if (real_dev) { 5795 ret = ethtool_get_ts_info_by_layer(real_dev, info); 5796 } else { 5797 /* Check if all slaves support software tx timestamping */ 5798 rcu_read_lock(); 5799 bond_for_each_slave_rcu(bond, slave, iter) { 5800 ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info); 5801 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) { 5802 sw_tx_support = true; 5803 continue; 5804 } 5805 5806 sw_tx_support = false; 5807 break; 5808 } 5809 rcu_read_unlock(); 5810 } 5811 5812 if (sw_tx_support) 5813 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE; 5814 5815 dev_put(real_dev); 5816 return ret; 5817 } 5818 5819 static const struct ethtool_ops bond_ethtool_ops = { 5820 .get_drvinfo = bond_ethtool_get_drvinfo, 5821 .get_link = ethtool_op_get_link, 5822 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5823 .get_ts_info = bond_ethtool_get_ts_info, 5824 }; 5825 5826 static const struct net_device_ops bond_netdev_ops = { 5827 .ndo_init = bond_init, 5828 .ndo_uninit = bond_uninit, 5829 .ndo_open = bond_open, 5830 .ndo_stop = bond_close, 5831 .ndo_start_xmit = bond_start_xmit, 5832 .ndo_select_queue = bond_select_queue, 5833 .ndo_get_stats64 = bond_get_stats, 5834 .ndo_eth_ioctl = bond_eth_ioctl, 5835 .ndo_siocbond = bond_do_ioctl, 5836 .ndo_siocdevprivate = bond_siocdevprivate, 5837 .ndo_change_rx_flags = bond_change_rx_flags, 5838 .ndo_set_rx_mode = bond_set_rx_mode, 5839 .ndo_change_mtu = bond_change_mtu, 5840 .ndo_set_mac_address = bond_set_mac_address, 5841 .ndo_neigh_setup = bond_neigh_setup, 5842 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5843 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5844 #ifdef CONFIG_NET_POLL_CONTROLLER 5845 .ndo_netpoll_setup = bond_netpoll_setup, 5846 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5847 .ndo_poll_controller = bond_poll_controller, 5848 #endif 5849 .ndo_add_slave = bond_enslave, 5850 .ndo_del_slave = bond_release, 5851 .ndo_fix_features = bond_fix_features, 5852 .ndo_features_check = passthru_features_check, 5853 .ndo_get_xmit_slave = bond_xmit_get_slave, 5854 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5855 .ndo_bpf = bond_xdp, 5856 .ndo_xdp_xmit = bond_xdp_xmit, 5857 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5858 .ndo_hwtstamp_get = bond_hwtstamp_get, 5859 .ndo_hwtstamp_set = bond_hwtstamp_set, 5860 }; 5861 5862 static const struct device_type bond_type = { 5863 .name = "bond", 5864 }; 5865 5866 static void bond_destructor(struct net_device *bond_dev) 5867 { 5868 struct bonding *bond = netdev_priv(bond_dev); 5869 5870 if (bond->wq) 5871 destroy_workqueue(bond->wq); 5872 5873 free_percpu(bond->rr_tx_counter); 5874 } 5875 5876 void bond_setup(struct net_device *bond_dev) 5877 { 5878 struct bonding *bond = netdev_priv(bond_dev); 5879 5880 spin_lock_init(&bond->mode_lock); 5881 bond->params = bonding_defaults; 5882 5883 /* Initialize pointers */ 5884 bond->dev = bond_dev; 5885 5886 /* Initialize the device entry points */ 5887 ether_setup(bond_dev); 5888 bond_dev->max_mtu = ETH_MAX_MTU; 5889 bond_dev->netdev_ops = &bond_netdev_ops; 5890 bond_dev->ethtool_ops = &bond_ethtool_ops; 5891 5892 bond_dev->needs_free_netdev = true; 5893 bond_dev->priv_destructor = bond_destructor; 5894 5895 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5896 5897 /* Initialize the device options */ 5898 bond_dev->flags |= IFF_MASTER; 5899 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5900 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5901 5902 #ifdef CONFIG_XFRM_OFFLOAD 5903 /* set up xfrm device ops (only supported in active-backup right now) */ 5904 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5905 INIT_LIST_HEAD(&bond->ipsec_list); 5906 mutex_init(&bond->ipsec_lock); 5907 #endif /* CONFIG_XFRM_OFFLOAD */ 5908 5909 /* don't acquire bond device's netif_tx_lock when transmitting */ 5910 bond_dev->lltx = true; 5911 5912 /* Don't allow bond devices to change network namespaces. */ 5913 bond_dev->netns_immutable = true; 5914 5915 /* By default, we declare the bond to be fully 5916 * VLAN hardware accelerated capable. Special 5917 * care is taken in the various xmit functions 5918 * when there are slaves that are not hw accel 5919 * capable 5920 */ 5921 5922 bond_dev->hw_features = MASTER_UPPER_DEV_VLAN_FEATURES | 5923 NETIF_F_HW_VLAN_CTAG_RX | 5924 NETIF_F_HW_VLAN_CTAG_FILTER | 5925 NETIF_F_HW_VLAN_STAG_RX | 5926 NETIF_F_HW_VLAN_STAG_FILTER; 5927 5928 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5929 bond_dev->features |= bond_dev->hw_features; 5930 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5931 bond_dev->features |= NETIF_F_GSO_PARTIAL; 5932 #ifdef CONFIG_XFRM_OFFLOAD 5933 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5934 /* Only enable XFRM features if this is an active-backup config */ 5935 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5936 bond_dev->features |= BOND_XFRM_FEATURES; 5937 #endif /* CONFIG_XFRM_OFFLOAD */ 5938 } 5939 5940 /* Destroy a bonding device. 5941 * Must be under rtnl_lock when this function is called. 5942 */ 5943 static void bond_uninit(struct net_device *bond_dev) 5944 { 5945 struct bonding *bond = netdev_priv(bond_dev); 5946 struct list_head *iter; 5947 struct slave *slave; 5948 5949 bond_netpoll_cleanup(bond_dev); 5950 5951 /* Release the bonded slaves */ 5952 bond_for_each_slave(bond, slave, iter) 5953 __bond_release_one(bond_dev, slave->dev, true, true); 5954 netdev_info(bond_dev, "Released all slaves\n"); 5955 5956 #ifdef CONFIG_XFRM_OFFLOAD 5957 mutex_destroy(&bond->ipsec_lock); 5958 #endif /* CONFIG_XFRM_OFFLOAD */ 5959 5960 bond_set_slave_arr(bond, NULL, NULL); 5961 5962 list_del_rcu(&bond->bond_list); 5963 5964 bond_debug_unregister(bond); 5965 } 5966 5967 /*------------------------- Module initialization ---------------------------*/ 5968 5969 static int __init bond_check_params(struct bond_params *params) 5970 { 5971 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 5972 struct bond_opt_value newval; 5973 const struct bond_opt_value *valptr; 5974 int arp_all_targets_value = 0; 5975 u16 ad_actor_sys_prio = 0; 5976 u16 ad_user_port_key = 0; 5977 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 5978 int arp_ip_count; 5979 int bond_mode = BOND_MODE_ROUNDROBIN; 5980 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 5981 int lacp_fast = 0; 5982 int tlb_dynamic_lb; 5983 5984 /* Convert string parameters. */ 5985 if (mode) { 5986 bond_opt_initstr(&newval, mode); 5987 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 5988 if (!valptr) { 5989 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 5990 return -EINVAL; 5991 } 5992 bond_mode = valptr->value; 5993 } 5994 5995 if (xmit_hash_policy) { 5996 if (bond_mode == BOND_MODE_ROUNDROBIN || 5997 bond_mode == BOND_MODE_ACTIVEBACKUP || 5998 bond_mode == BOND_MODE_BROADCAST) { 5999 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 6000 bond_mode_name(bond_mode)); 6001 } else { 6002 bond_opt_initstr(&newval, xmit_hash_policy); 6003 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 6004 &newval); 6005 if (!valptr) { 6006 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 6007 xmit_hash_policy); 6008 return -EINVAL; 6009 } 6010 xmit_hashtype = valptr->value; 6011 } 6012 } 6013 6014 if (lacp_rate) { 6015 if (bond_mode != BOND_MODE_8023AD) { 6016 pr_info("lacp_rate param is irrelevant in mode %s\n", 6017 bond_mode_name(bond_mode)); 6018 } else { 6019 bond_opt_initstr(&newval, lacp_rate); 6020 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 6021 &newval); 6022 if (!valptr) { 6023 pr_err("Error: Invalid lacp rate \"%s\"\n", 6024 lacp_rate); 6025 return -EINVAL; 6026 } 6027 lacp_fast = valptr->value; 6028 } 6029 } 6030 6031 if (ad_select) { 6032 bond_opt_initstr(&newval, ad_select); 6033 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 6034 &newval); 6035 if (!valptr) { 6036 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 6037 return -EINVAL; 6038 } 6039 params->ad_select = valptr->value; 6040 if (bond_mode != BOND_MODE_8023AD) 6041 pr_warn("ad_select param only affects 802.3ad mode\n"); 6042 } else { 6043 params->ad_select = BOND_AD_STABLE; 6044 } 6045 6046 if (max_bonds < 0) { 6047 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 6048 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 6049 max_bonds = BOND_DEFAULT_MAX_BONDS; 6050 } 6051 6052 if (miimon < 0) { 6053 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6054 miimon, INT_MAX); 6055 miimon = 0; 6056 } 6057 6058 if (updelay < 0) { 6059 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6060 updelay, INT_MAX); 6061 updelay = 0; 6062 } 6063 6064 if (downdelay < 0) { 6065 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6066 downdelay, INT_MAX); 6067 downdelay = 0; 6068 } 6069 6070 if (use_carrier != 1) { 6071 pr_err("Error: invalid use_carrier parameter (%d)\n", 6072 use_carrier); 6073 return -EINVAL; 6074 } 6075 6076 if (num_peer_notif < 0 || num_peer_notif > 255) { 6077 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 6078 num_peer_notif); 6079 num_peer_notif = 1; 6080 } 6081 6082 /* reset values for 802.3ad/TLB/ALB */ 6083 if (!bond_mode_uses_arp(bond_mode)) { 6084 if (!miimon) { 6085 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n"); 6086 pr_warn("Forcing miimon to 100msec\n"); 6087 miimon = BOND_DEFAULT_MIIMON; 6088 } 6089 } 6090 6091 if (tx_queues < 1 || tx_queues > 255) { 6092 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 6093 tx_queues, BOND_DEFAULT_TX_QUEUES); 6094 tx_queues = BOND_DEFAULT_TX_QUEUES; 6095 } 6096 6097 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 6098 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 6099 all_slaves_active); 6100 all_slaves_active = 0; 6101 } 6102 6103 if (resend_igmp < 0 || resend_igmp > 255) { 6104 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 6105 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 6106 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 6107 } 6108 6109 bond_opt_initval(&newval, packets_per_slave); 6110 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 6111 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 6112 packets_per_slave, USHRT_MAX); 6113 packets_per_slave = 1; 6114 } 6115 6116 if (bond_mode == BOND_MODE_ALB) { 6117 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n", 6118 updelay); 6119 } 6120 6121 if (!miimon) { 6122 if (updelay || downdelay) { 6123 /* just warn the user the up/down delay will have 6124 * no effect since miimon is zero... 6125 */ 6126 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n", 6127 updelay, downdelay); 6128 } 6129 } else { 6130 /* don't allow arp monitoring */ 6131 if (arp_interval) { 6132 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6133 miimon, arp_interval); 6134 arp_interval = 0; 6135 } 6136 6137 if ((updelay % miimon) != 0) { 6138 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6139 updelay, miimon, (updelay / miimon) * miimon); 6140 } 6141 6142 updelay /= miimon; 6143 6144 if ((downdelay % miimon) != 0) { 6145 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6146 downdelay, miimon, 6147 (downdelay / miimon) * miimon); 6148 } 6149 6150 downdelay /= miimon; 6151 } 6152 6153 if (arp_interval < 0) { 6154 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6155 arp_interval, INT_MAX); 6156 arp_interval = 0; 6157 } 6158 6159 for (arp_ip_count = 0, i = 0; 6160 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6161 __be32 ip; 6162 6163 /* not a complete check, but good enough to catch mistakes */ 6164 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6165 !bond_is_ip_target_ok(ip)) { 6166 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6167 arp_ip_target[i]); 6168 arp_interval = 0; 6169 } else { 6170 if (bond_get_targets_ip(arp_target, ip) == -1) 6171 arp_target[arp_ip_count++] = ip; 6172 else 6173 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6174 &ip); 6175 } 6176 } 6177 6178 if (arp_interval && !arp_ip_count) { 6179 /* don't allow arping if no arp_ip_target given... */ 6180 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6181 arp_interval); 6182 arp_interval = 0; 6183 } 6184 6185 if (arp_validate) { 6186 if (!arp_interval) { 6187 pr_err("arp_validate requires arp_interval\n"); 6188 return -EINVAL; 6189 } 6190 6191 bond_opt_initstr(&newval, arp_validate); 6192 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6193 &newval); 6194 if (!valptr) { 6195 pr_err("Error: invalid arp_validate \"%s\"\n", 6196 arp_validate); 6197 return -EINVAL; 6198 } 6199 arp_validate_value = valptr->value; 6200 } else { 6201 arp_validate_value = 0; 6202 } 6203 6204 if (arp_all_targets) { 6205 bond_opt_initstr(&newval, arp_all_targets); 6206 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6207 &newval); 6208 if (!valptr) { 6209 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6210 arp_all_targets); 6211 arp_all_targets_value = 0; 6212 } else { 6213 arp_all_targets_value = valptr->value; 6214 } 6215 } 6216 6217 if (miimon) { 6218 pr_info("MII link monitoring set to %d ms\n", miimon); 6219 } else if (arp_interval) { 6220 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6221 arp_validate_value); 6222 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6223 arp_interval, valptr->string, arp_ip_count); 6224 6225 for (i = 0; i < arp_ip_count; i++) 6226 pr_cont(" %s", arp_ip_target[i]); 6227 6228 pr_cont("\n"); 6229 6230 } else if (max_bonds) { 6231 /* miimon and arp_interval not set, we need one so things 6232 * work as expected, see bonding.txt for details 6233 */ 6234 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n"); 6235 } 6236 6237 if (primary && !bond_mode_uses_primary(bond_mode)) { 6238 /* currently, using a primary only makes sense 6239 * in active backup, TLB or ALB modes 6240 */ 6241 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6242 primary, bond_mode_name(bond_mode)); 6243 primary = NULL; 6244 } 6245 6246 if (primary && primary_reselect) { 6247 bond_opt_initstr(&newval, primary_reselect); 6248 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6249 &newval); 6250 if (!valptr) { 6251 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6252 primary_reselect); 6253 return -EINVAL; 6254 } 6255 primary_reselect_value = valptr->value; 6256 } else { 6257 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6258 } 6259 6260 if (fail_over_mac) { 6261 bond_opt_initstr(&newval, fail_over_mac); 6262 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6263 &newval); 6264 if (!valptr) { 6265 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6266 fail_over_mac); 6267 return -EINVAL; 6268 } 6269 fail_over_mac_value = valptr->value; 6270 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6271 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6272 } else { 6273 fail_over_mac_value = BOND_FOM_NONE; 6274 } 6275 6276 bond_opt_initstr(&newval, "default"); 6277 valptr = bond_opt_parse( 6278 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6279 &newval); 6280 if (!valptr) { 6281 pr_err("Error: No ad_actor_sys_prio default value"); 6282 return -EINVAL; 6283 } 6284 ad_actor_sys_prio = valptr->value; 6285 6286 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6287 &newval); 6288 if (!valptr) { 6289 pr_err("Error: No ad_user_port_key default value"); 6290 return -EINVAL; 6291 } 6292 ad_user_port_key = valptr->value; 6293 6294 bond_opt_initstr(&newval, "default"); 6295 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6296 if (!valptr) { 6297 pr_err("Error: No tlb_dynamic_lb default value"); 6298 return -EINVAL; 6299 } 6300 tlb_dynamic_lb = valptr->value; 6301 6302 if (lp_interval == 0) { 6303 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6304 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6305 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6306 } 6307 6308 /* fill params struct with the proper values */ 6309 params->mode = bond_mode; 6310 params->xmit_policy = xmit_hashtype; 6311 params->miimon = miimon; 6312 params->num_peer_notif = num_peer_notif; 6313 params->arp_interval = arp_interval; 6314 params->arp_validate = arp_validate_value; 6315 params->arp_all_targets = arp_all_targets_value; 6316 params->missed_max = 2; 6317 params->updelay = updelay; 6318 params->downdelay = downdelay; 6319 params->peer_notif_delay = 0; 6320 params->lacp_active = 1; 6321 params->lacp_fast = lacp_fast; 6322 params->primary[0] = 0; 6323 params->primary_reselect = primary_reselect_value; 6324 params->fail_over_mac = fail_over_mac_value; 6325 params->tx_queues = tx_queues; 6326 params->all_slaves_active = all_slaves_active; 6327 params->resend_igmp = resend_igmp; 6328 params->min_links = min_links; 6329 params->lp_interval = lp_interval; 6330 params->packets_per_slave = packets_per_slave; 6331 params->tlb_dynamic_lb = tlb_dynamic_lb; 6332 params->ad_actor_sys_prio = ad_actor_sys_prio; 6333 eth_zero_addr(params->ad_actor_system); 6334 params->ad_user_port_key = ad_user_port_key; 6335 params->coupled_control = 1; 6336 params->broadcast_neighbor = 0; 6337 if (packets_per_slave > 0) { 6338 params->reciprocal_packets_per_slave = 6339 reciprocal_value(packets_per_slave); 6340 } else { 6341 /* reciprocal_packets_per_slave is unused if 6342 * packets_per_slave is 0 or 1, just initialize it 6343 */ 6344 params->reciprocal_packets_per_slave = 6345 (struct reciprocal_value) { 0 }; 6346 } 6347 6348 if (primary) 6349 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6350 6351 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6352 #if IS_ENABLED(CONFIG_IPV6) 6353 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6354 #endif 6355 6356 return 0; 6357 } 6358 6359 /* Called from registration process */ 6360 static int bond_init(struct net_device *bond_dev) 6361 { 6362 struct bonding *bond = netdev_priv(bond_dev); 6363 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6364 6365 netdev_dbg(bond_dev, "Begin bond_init\n"); 6366 6367 bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM, 6368 bond_dev->name); 6369 if (!bond->wq) 6370 return -ENOMEM; 6371 6372 bond->notifier_ctx = false; 6373 6374 spin_lock_init(&bond->stats_lock); 6375 netdev_lockdep_set_classes(bond_dev); 6376 6377 list_add_tail_rcu(&bond->bond_list, &bn->dev_list); 6378 6379 bond_prepare_sysfs_group(bond); 6380 6381 bond_debug_register(bond); 6382 6383 /* Ensure valid dev_addr */ 6384 if (is_zero_ether_addr(bond_dev->dev_addr) && 6385 bond_dev->addr_assign_type == NET_ADDR_PERM) 6386 eth_hw_addr_random(bond_dev); 6387 6388 return 0; 6389 } 6390 6391 unsigned int bond_get_num_tx_queues(void) 6392 { 6393 return tx_queues; 6394 } 6395 6396 /* Create a new bond based on the specified name and bonding parameters. 6397 * If name is NULL, obtain a suitable "bond%d" name for us. 6398 * Caller must NOT hold rtnl_lock; we need to release it here before we 6399 * set up our sysfs entries. 6400 */ 6401 int bond_create(struct net *net, const char *name) 6402 { 6403 struct net_device *bond_dev; 6404 struct bonding *bond; 6405 int res = -ENOMEM; 6406 6407 rtnl_lock(); 6408 6409 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6410 name ? name : "bond%d", NET_NAME_UNKNOWN, 6411 bond_setup, tx_queues); 6412 if (!bond_dev) 6413 goto out; 6414 6415 bond = netdev_priv(bond_dev); 6416 dev_net_set(bond_dev, net); 6417 bond_dev->rtnl_link_ops = &bond_link_ops; 6418 6419 res = register_netdevice(bond_dev); 6420 if (res < 0) { 6421 free_netdev(bond_dev); 6422 goto out; 6423 } 6424 6425 netif_carrier_off(bond_dev); 6426 6427 bond_work_init_all(bond); 6428 6429 out: 6430 rtnl_unlock(); 6431 return res; 6432 } 6433 6434 static int __net_init bond_net_init(struct net *net) 6435 { 6436 struct bond_net *bn = net_generic(net, bond_net_id); 6437 6438 bn->net = net; 6439 INIT_LIST_HEAD(&bn->dev_list); 6440 6441 bond_create_proc_dir(bn); 6442 bond_create_sysfs(bn); 6443 6444 return 0; 6445 } 6446 6447 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters 6448 * race condition in bond unloading") we need to remove sysfs files 6449 * before we remove our devices (done later in bond_net_exit_rtnl()) 6450 */ 6451 static void __net_exit bond_net_pre_exit(struct net *net) 6452 { 6453 struct bond_net *bn = net_generic(net, bond_net_id); 6454 6455 bond_destroy_sysfs(bn); 6456 } 6457 6458 static void __net_exit bond_net_exit_rtnl(struct net *net, 6459 struct list_head *dev_kill_list) 6460 { 6461 struct bond_net *bn = net_generic(net, bond_net_id); 6462 struct bonding *bond, *tmp_bond; 6463 6464 /* Kill off any bonds created after unregistering bond rtnl ops */ 6465 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6466 unregister_netdevice_queue(bond->dev, dev_kill_list); 6467 } 6468 6469 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory 6470 * only after all bonds are gone") bond_destroy_proc_dir() is called 6471 * after bond_net_exit_rtnl() has completed. 6472 */ 6473 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6474 { 6475 struct bond_net *bn; 6476 struct net *net; 6477 6478 list_for_each_entry(net, net_list, exit_list) { 6479 bn = net_generic(net, bond_net_id); 6480 bond_destroy_proc_dir(bn); 6481 } 6482 } 6483 6484 static struct pernet_operations bond_net_ops = { 6485 .init = bond_net_init, 6486 .pre_exit = bond_net_pre_exit, 6487 .exit_rtnl = bond_net_exit_rtnl, 6488 .exit_batch = bond_net_exit_batch, 6489 .id = &bond_net_id, 6490 .size = sizeof(struct bond_net), 6491 }; 6492 6493 static int __init bonding_init(void) 6494 { 6495 int i; 6496 int res; 6497 6498 res = bond_check_params(&bonding_defaults); 6499 if (res) 6500 goto out; 6501 6502 bond_create_debugfs(); 6503 6504 res = register_pernet_subsys(&bond_net_ops); 6505 if (res) 6506 goto err_net_ops; 6507 6508 res = bond_netlink_init(); 6509 if (res) 6510 goto err_link; 6511 6512 for (i = 0; i < max_bonds; i++) { 6513 res = bond_create(&init_net, NULL); 6514 if (res) 6515 goto err; 6516 } 6517 6518 skb_flow_dissector_init(&flow_keys_bonding, 6519 flow_keys_bonding_keys, 6520 ARRAY_SIZE(flow_keys_bonding_keys)); 6521 6522 register_netdevice_notifier(&bond_netdev_notifier); 6523 out: 6524 return res; 6525 err: 6526 bond_netlink_fini(); 6527 err_link: 6528 unregister_pernet_subsys(&bond_net_ops); 6529 err_net_ops: 6530 bond_destroy_debugfs(); 6531 goto out; 6532 6533 } 6534 6535 static void __exit bonding_exit(void) 6536 { 6537 unregister_netdevice_notifier(&bond_netdev_notifier); 6538 6539 bond_netlink_fini(); 6540 unregister_pernet_subsys(&bond_net_ops); 6541 6542 bond_destroy_debugfs(); 6543 6544 #ifdef CONFIG_NET_POLL_CONTROLLER 6545 /* Make sure we don't have an imbalance on our netpoll blocking */ 6546 WARN_ON(atomic_read(&netpoll_block_tx)); 6547 #endif 6548 } 6549 6550 module_init(bonding_init); 6551 module_exit(bonding_exit); 6552 MODULE_LICENSE("GPL"); 6553 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6554 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6555 MODULE_IMPORT_NS("NETDEV_INTERNAL"); 6556